CN103824085B - Image processing, image recognition and image classification apparatus and method - Google Patents

Image processing, image recognition and image classification apparatus and method Download PDF

Info

Publication number
CN103824085B
CN103824085B CN201410102668.7A CN201410102668A CN103824085B CN 103824085 B CN103824085 B CN 103824085B CN 201410102668 A CN201410102668 A CN 201410102668A CN 103824085 B CN103824085 B CN 103824085B
Authority
CN
China
Prior art keywords
target
candidate
layer
object
image
Prior art date
Application number
CN201410102668.7A
Other languages
Chinese (zh)
Other versions
CN103824085A (en
Inventor
门洪涛
Original Assignee
苏州比特速浪电子科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 苏州比特速浪电子科技有限公司 filed Critical 苏州比特速浪电子科技有限公司
Priority to CN201410102668.7A priority Critical patent/CN103824085B/en
Publication of CN103824085A publication Critical patent/CN103824085A/en
Application granted granted Critical
Publication of CN103824085B publication Critical patent/CN103824085B/en

Links

Abstract

本发明公开了一种图像处理、图像识别及图像分类的装置及其方法,对于从拍摄的未特定对象物的写真等未特定图像数据提取的目标,进行必要的变换,可提取对应已知对象物的新目标。 The present invention discloses an image processing apparatus and method for image recognition and image classification, for non-specific extraction target image data of images photographed is not the specific object from the other, the necessary conversion, may extract the corresponding known object the new target. 相似度测定单元读取相似度判定程序,进行初期候补目标的提取。 Similarity measuring unit reads the degree of similarity determination program, the initial candidate object extraction. 候补目标变换单元读取增厚·减薄用目标选择程序、候补目标增厚·减薄程序,变换初期候补目标。 Conversion candidate target cell read-thinning thickened with the target selection process, the target candidate-thinning thickening procedures, the initial candidate target conversion. 进而,目标确定单元对变换的目标进行相似度测定,根据需要,重复变换动作,确定应该提取的目标。 Further, the target determination unit similarity determination target transformed, if necessary, repeat the conversion operation, it is determined to be the target of extraction. 以此,可提取对应已知对象物的新目标。 In this, the new target corresponding to a known object can be extracted.

Description

图像处理、图像识别及图像分类的装置及其方法 Image processing, image recognition and image classification apparatus and method

技术领域 FIELD

[0001]本发明涉及一种图像处理、图像识别及图像分类的装置及其方法,其通过从未知对象物的图像数据提取目标进行变换,获得已知对象物的新目标。 [0001] The present invention relates to apparatus and method for image processing, image recognition and image classification which extracts the target image data is converted from an unknown object to obtain a new target object is known.

背景技术 Background technique

[0002]目前,在对图像进行处理时,通过电脑等图像处理装置的处理,从拍摄的未知对象物即不明对象物的写真等图像数据提取图像时,并不是刚刚恰如其分的提取了对应的已知对象物部分。 [0002] Currently, when the image processing by the image processing apparatus like computers, the unknown object from photographed image extraction i.e., not just unknown appropriately extracting object images and the like corresponding to the image data has been known object portions. 比如,在图像处理过程中,作为本应被提取的部分的一部分,即应该被包含的图像数据的部分图像被删除,或者恰恰相反,本应是不需要的一部分,即应该被删除的图像数据的部分图像反而又被包含。 For example, in the image processing, the partial image to be part of the present image portion is extracted, i.e., should be included in the data is deleted, or on the contrary, it is not required to be part of the present, i.e., image data that should be deleted the part of the image but has been included. 如果上述过程是人眼根据图像数据进行提取,基于进行提取的人的经验、知识,贴切地判断必要部分,可提取恰当的所需图像数据。 If the above procedure extracted human eye image data, based on the extract of human experience, knowledge, aptly judged necessary portions, a desired image can be extracted data appropriately. 然而,通过这样的人类判断提取的范围的判断,是比较抽象的,概念广、而且具有多样性的,但在通过电脑等控制的图像处理装置中,也进行与人类同样的判断,并非易事。 However, this is determined by the extracted range of human judgment, abstract is relatively wide concept, but also diversity, but by the image processing device control computers and the like, and subjected to the same human judgment, easy .

[0003]比如,根据一般写真高精度提取图像内目标(部分图像)的技术,关于被分为多个階调的边缘图像(階调边缘图像),对应与属于1个階调的階调边缘图像相邻的属于其他階调的階调边缘图像的领域数,决定是否提取图像内目标的知识被广为人知(参考专利文献1,特开平11-25271号公报)。 [0003] For example, to extract the image of a target with high accuracy according to the general images (partial images) technique, on an edge image is divided into a plurality of gradation (gradation edge image), belonging to a corresponding tone gradation edge knowledge of the number of images belonging to other areas adjacent to the tone gradation of the edge image is determined whether to extract the image of the object is known (refer to Patent Document 1, Japanese Patent Publication No. 11-25271). 但是,在专利文献1的情况下,因为图像的边缘部分等作为是否提取图像内目标的判断基准,对于仅仅依靠图像的边缘信息,很难判断是否是应该提取的对象物,因此,在提取部分图像时,并不能恰当的提取图像内的目标。 However, in the case of Patent Document 1, since the edge portion extracted as an image is determined whether the reference image of the target, for the edge information of the image alone, is difficult to determine whether the object should be extracted, and therefore, the extraction section when the image is not an appropriate target in the image is extracted.

发明内容 SUMMARY

[0004]为了克服上述缺陷,本发明提供了一种图像处理、图像识别及图像分类的装置及其方法,根据拍摄的未知对象物的写真等未知图像提取的目标进行必要的变换,从而提取对应已知对象物的新目标。 [0004] In order to overcome the above drawbacks, the present invention provides an image processing apparatus and method for image recognition and image classification, the necessary conversion based on the target image extracted unknown unknown object captured images and the like, thereby extracting the corresponding known new target object. 在此,所谓目标,是指具有一定连贯的部分图像,由对应此图像部分的一定范围领域内的一群像素信息构成。 Here, the target, is a portion having a certain coherent image, composed by a group of the pixel information corresponding to a range within a certain field of the image portion.

[0005]本发明为了解决其技术问题所采用的技术方案是: [0005] Technical Solution In order to solve the technical problem is that:

[0006]本发明所述的图像处理装置是根据拍摄的未知对象物的未知图像数据所提取的目标,进行变换后提取对应已知对象物的新目标的图像处理装置。 The image processing apparatus according to [0006] of the present invention is the unknown image data of the unknown object extracted target captured, the image processing apparatus known to the new target object after extracting the corresponding transformed. 该图像处理装置包含以下单元:生成层的层生成单元,该层由包含根据未知图像数据提取的目标的分割图像集合体构成;决定初期候补目标的初期候补目标决定单元,即根据层生成单元生成的层中包含的目标,进行变换而得到作为初期候补的初期候补目标;候补目标变换单元,即利用层生成单元生成的层中包含的初期候补目标以外的目标,变换初期候补目标,生成新的候补目标的候补目标变换单元。 The image processing apparatus comprises the following elements: generation layer layer generating means, which layer consists of an aggregate of divided images comprising object extraction based on the unknown image data; initial candidate target determining means determines the initial candidate object, according to the layer generation unit which generates object contained in the layer, obtained by converting as the initial candidate initial candidate target; candidate target transform unit, i.e. with the target than the initial candidate target layer layer generating unit included, converting the initial candidate target to generate a new candidate target candidate object transformation unit.

[0007]通过上述图像处理装置,在候补目标变换单元,变换初期候补目标,生成新的候补目标。 [0007] By the above-described image processing apparatus, the target candidate converting means converts the initial target candidates, generating a new candidate targets. 即对于根据拍摄的未知对象物的写真等未知图像数据提取的目标进行必要的变换, 可提取对应已知对象物的新目标。 That is necessary for converting the extracted target object based on images captured unknown unknown image data and the like, may be extracted corresponding to the new target object is known.

[0008]本发明的具体侧面,还具有以下单元:测定目标与已知对象物相似度的相似度测定单元,以及将基于相似度测定单元测定的相似度测定结果,通过候补目标变换单元的变换而生成的新候补目标和从变换前的候补目标中选择的目标作为确定目标进行保存的目标确定单元。 [0008] The particular aspect of the present invention further includes the following units: a measurement unit measuring the degree of similarity with known target object similarity, and the similarity measurement based on the measurement result of the similarity determination unit, by converting means converting the target candidate newly generated and the target and the target candidate selecting from the candidate target before conversion stored in the target as a determination target determination unit. 此时,基于相似度的测定,对于被变换的目标,可以更加确切的确定是否为应该提取的目标。 At this point, based on the determination of similarity, the target was transformed, we can more precisely determine whether the target should be extracted.

[0009]本发明的另一侧面,初期候补目标决定单元基于相似度测定单元的测定结果,来决定初期候补目标。 [0009] Another aspect of the present invention, the initial candidate target determining unit based on the measurement result of the similarity determination unit to determine the initial target candidates. 此时,可促进应该提取的目标的早期决定。 At this point, can promote early decision should be extracted goals.

[0010]本发明另一侧面,目标确定单元基于相似度测定单元测定的相似度时事先规定的阈值,决定作为确定目标应该被保存的目标。 [0010] Another aspect of the present invention, the target value determination threshold when the degree of similarity measurement unit based on the similarity measurement unit specified in advance, determines a target should be saved as the determination target. 此时,比如,对于超过事先规定的阈值的目标, 所部作为确定目标进行保存,若作为被提取对象,可降低遗漏应该被提取的目标,未进行提取的可能性。 In this case, for example, the target threshold value exceeds a predetermined advance, to save his troops carried out as a determination target, as if the object is extracted, can be omitted to reduce the target to be extracted, the possibility of extraction is not performed. (此时,比如,对于超过事先规定的阈值的目标,将作为确定目标进行保存,可降低遗漏应该被提取的目标,未进行提取的可能性。) (In this case, for example, the target threshold value exceeds a predetermined advance, will be stored as the determination target, the target may be missed should be extracted to reduce the possibility of extraction is not performed.)

[0011]本发明另一侧面,目标确定单元对基于相似度测定单元测定的相似度测定结果, 通过候补目标变换单元的变换而生成的新候补目标和变换前的候补目标进行比较,判断是该重复该候补目标变换单元的变换动作还是停止该动作,将判断结果即得到的候补目标作为确定目标进行保存。 [0011] Another aspect of the present invention, the target determining unit based on the similarity measurement unit measures the measurement result of the degree of similarity, the new candidate objects and candidate target generated before conversion by the conversion means converting the target candidate, and determines that the the candidate target transformation unit repeats transformation or stop the operation of the operation, i.e., the determination result obtained by the candidate target stored as determination target. 此时,通过变换动作,可提取被认为最恰当的目标。 At this time, the converting operation may be considered the most appropriate extraction target.

[0012]本发明的另一侧面,初期候补目标决定单元将层生成单元生成的各种目标中,超过相似度测定单元测定的相似度阈值的目标作为初期候补目标。 [0012] Another aspect of the present invention, the initial candidate target determining unit layer generating unit generates the various objects, the similarity of objects similarity threshold measured by the measuring unit exceeds a certain initial candidate. 此时,通过选择相似度较高的目标作为初期候补目标,可控制变换处理的计算量。 In this case, by selecting a high similarity as an initial target a target candidate, transformation processing calculation amount can be controlled.

[0013]本发明的另一侧面,相似度测定单元针对已知对象物进行相似度测定,将提取的目标与已知对象物近似性的高低进行数值化的相似度(确信度)作为测定的基准。 [0013] Another aspect of the present invention, a similarity measurement unit for measuring the similarity of the object is known, and the level of the extracted target object proximity known numerically similarity (degree of certainty) as determined benchmark. 此时,基于相似度测定单元的测定,根据数值比较,可判断近似性的高低。 At this time, the measurement unit based on the similarity measurement, in accordance with the comparison value, can determine the level of similarity.

[0014]本发明另一侧面,层生成单元具有第1层生成单元即生成由包含初期候补目标的目标群构成的候补层和第2层生成单元即生成由不包含初期候补目标的目标群构成的特别层。 [0014] Another aspect of the present invention, a layer having a layer generating unit generating means which generates a first layer and a second candidate generating unit composed of the initial candidate target comprises a target group, i.e. generated by the target group does not include the initial candidate objects constituting special layer. 第2层生成单元生成特别层,即生成由附带目标构成的附带层,该附带目标在根据未知图像数据提取目标时的图像处理中从构成候补层的各种目标被排除,而在候补目标变换单元的变换中,附带于其他目标。 Layer generating unit generates special layer 2, i.e., comes generating layer composed of incidental target, the target comes when extracting the target image processing according to the unknown image data are excluded from the target of various layers constituting the candidate, and the candidate target transform conversion unit, with the other goals. 此时,比如,决定候补层时,即使有关于缺少部分图像的信息,也会被作为附带目标,被作为最终确定目标的一部分。 At this time, for example, the candidate decided layer, even if the partial image information is missing, will be included as a target, a target is determined as a final part.

[0015]本发明的另一侧面,层生成单元还具有第3层生成单元即生成根据已知对象物及别的己知对象物提取的其他候补目标相关分割图像的集合体构成的其他候补层,候补目标变换单元在生成新的候补目标时,将包含于其他候补层的其他候补目标从变换动作对象中排除。 [0015] Another aspect of the present invention, another layer generating unit further has a third layer Layer candidate generating means which generates divided images related to an assembly of a target according to other known candidate object and others known in the object extraction other candidate target, the target candidate converting means when generating a new candidate targets contained in the other layers is excluded from the candidate object conversion operation. 此时,可以事先集中变换处理对象。 In this case, the object may be previously concentrated conversion process.

[0016]本发明的另一侧面,层生成单元生成的层像包含相对处于上位的层和相对处于下位的层的阶层性构造的层群一样。 [0016] Another aspect of the present invention, layer generation unit comprises the same layer as an upper layer group in the opposite configuration relative hierarchical layer and in the lower layer. 上位侧的层由父目标构成,父目标将构成下位侧层的多个目标作为子目标包含在内部。 An upper layer composed of a parent object side, a plurality of parent object constituting the lower side of the target as a sub-target layer contained therein. 此时,因为有阶层性构造,整理变换处理的选择项,可顺利进行处理。 In this case, because there is a hierarchical structure, finishing options transform processing can be handled smoothly.

[0017] 本发明的另一侧面,初期候补目标决定单元在层生成单元生成的阶层性构造的层群中,根据下位侧层的目标,决定初期候补目标。 [0017] Another aspect of the present invention, the initial candidate decision unit in the target layer, layer group generation unit in the hierarchical structure, the lower side of the target layer, the target determines the initial candidate. 候补目标变换单元对于初期候补目标,使之附带其他目标,生成新的候补目标。 Means for converting the initial candidate target candidate target, other targets so that it comes to generate a new candidate targets. 此时,通过从下位侧层到上位侧层自下而上,可提取目的目标。 In this case, the bottom-side from the lower layer to the upper layer side, the target object may be extracted.

[0018]本发明的另一侧面,初期候补目标决定单元在层生成单元生成的阶层性构造的层群中,从属于上位侧的目标中决定初期候补目标,候补目标变换单元针对初期候补目标,去除一部分被包含于初期候补目标中的子目标,生成新的候补目标。 [0018] Another aspect of the present invention, the initial candidate decision unit in the target layer, layer group generation unit in the hierarchical structure, belonging to the upper side of the target candidates determined initial target, the target candidate for a target candidate initial transformation unit, removing a portion of the child object is contained in the initial candidate target to generate a new candidate targets. 此时,从上位侧层到下位侧层自上而下,可提取目的目标。 At this time, from the upper side to the lower side of the layer from top to bottom layer, the target object may be extracted.

[0019]本发明另一侧面,层生成单元在既存分割图像中,测定各分割图像的尺寸,比较被测定的该部分分割图像的图像数据大小和对应已知对象物的图像数据大小,根据需要,对该部分分割图像进行再分割,生成基于新目标的层。 [0019] Another aspect of the present invention, in the existing layer generating means divided images, determining the size of each divided image, the image data size of the divided image portion and comparing the measured object corresponding to the known size of image data, as needed , the divided image portions subdividing the target based on the new generation layer. 此时,比如,可以避免理应被提取的目标被埋在大块目标中而被漏看的情况。 In this case, for example, should be avoided where the extracted target buried in the bulk of the target is missed in.

[0020]为达成上述目的,本发明变换根据拍摄的未知对象物的未知图像数据提取的目标,提取对应已知对象物的新目标的图像处理方法。 [0020] To achieve the above object, the present invention transform the target image data of the unknown object captured unknown extracted, extracting an image processing method corresponding to the new target object is known. 具有以下工序:(1)层生成工序,生成由包含根据未知图像数据提取的目标的分割图像集合体构成的层;(2)初期候补目标决定工序,根据被包含于层生成工序生成的层中的目标,决定初期候补目标即作为初期候补而被变换所得处理对象;(3)候补目标变换工序,利用被包含于层生成工序生成的层中的初期候补目标以外的目标,变换初期候补目标,生成新的初期候补目标。 Having the following steps: (1) layer generating step generates the layers comprising the aggregate of divided image data of the unknown image composed of the extracted object; (2) certain initial candidate decision step, according to a layer comprising a layer generating step is generated target, determines the initial candidate target that is defined as the initial candidate is converted resultant processed; (3) a candidate target conversion step, using the target other than the initial candidate objects are contained in the layer generating step of generating in the layer, transform the initial candidate target, the new generation of the initial target candidates.

[0021]通过上述图像处理方法,基于相似度测定工序的相似度,进行最初候补即初期候补目标的提取,进而,在候补目标变换工序,不断变换初期候补目标,测定被变换目标的相似度,确定应该被提取的目标。 [0021] By the above-described image processing method, based on the similarity of the similarity determination step, the first candidate for the initial candidate object extraction i.e., Further, in the step of converting the candidate target, changing the initial target candidates determined by the similarity transformation of the object, targeting should be extracted. 以此,通过对根据所拍摄未知对象物的写真等未知图像数据提取的目标进行必要变换,可提取对应已知对象物的新目标。 In this, performs necessary conversion by the target object images and other unknown unknown captured image data extracted, can extract a new object corresponding to a known target.

[0022]为达成上述目的,本发明相关图像识别装置是根据所拍摄未知对象物的未知图像数据,进行已知对象物相关识别的图像识别装置。 [0022] To achieve the above object, the present invention is related to an image recognition apparatus according to the unknown unknown object captured image data, the image recognition means known object-related identification. 包含上述任何一项记载的图像处理装置, 根据该图像处理装置,基于根据上述未知图像数据提取的新目标,进行有无已知对象物的判定。 The image processing device includes any of the above described, the image processing apparatus, based on the new unknown object extraction based on the image data, determines the presence or absence of a known object.

[0023]通过上述图像识别装置,贴切地提取对应已知对象物的新目标,可进行更确切的图像识别。 [0023] by the image recognition means, apt to extract the new target object corresponding to the known objects may be made more precise image recognition.

[0024]为达成上述目的,本发明所述图像识别方法是根据所拍摄未知对象物的未知图像数据,进行己知对象物相关识别的图像识别方法。 [0024] To achieve the above object, the image recognition method of the present invention is the unknown captured image data of the unknown object, an image recognition method known related object recognition. 具有判定工序,即包含上述记载的图像图像处理方法,根据该图像处理方法,基于根据上述未知图像数据提取的新目标,进行已知对象物有无的判定。 Having a determination step of, i.e., comprising the image processing method described above, according to the image processing method, based on a new unknown object extraction based on the image data, determines the presence or absence of an object is known.

[0025]为达成上述目的,本发明所述图像处理装置根据所拍摄未知对象物的未知图像数据,进行已知对象物相关识别的图像分类装置。 [0025] To achieve the above object, the present invention is an image processing apparatus according to the unknown unknown object captured image data, the image classification means known object-related identification. 包含上述任何一项记载的图像处理装置,根据该图像处理装置,基于根据上述未知图像数据提取的新目标,进行已知对象物的分类。 The image processing device includes any of the above described, the image processing apparatus, based on the new unknown object extraction based on the image data, to classify the object is known. [0026]根据上述图像分类装置,贴切地提取对应已知对象物的新目标,可更加确切的进行图像分类。 [0026] According to the above-described image classification means, corresponding to the new target extraction aptly known object, an image can be more precise classification.

[0027]为达成上述目的,本发明所述分类方法是根据所拍摄未知对象物的未知图像数据,进行已知对象物相关识别的图像分类方法。 [0027] To achieve the above object, the classification method of the present invention is the unknown unknown object captured image data, an image classification method known in the relevant object recognition. 具有分类工序,即包含上述记载的图像处理方法,根据该图像处理方法,基于根据上述未知图像数据提取的新目标,进行已知对象物的分类。 Having a classification step, i.e. including the image processing method described above, according to the image processing method, based on a new unknown object extraction based on the image data, to classify the object is known.

[0028]本发明的有益效果是:本发明所述图像处理、图像识别及图像分类的装置及其方法,提供了一种新的图像处理、图像识别和分类的装置和方法,能够相对人眼更加快捷准确从未知对象物的未知图像数据判定有无已知对象物。 [0028] Advantageous effects of the present invention are: the present invention, an image processing apparatus and method for image recognition and image classification, there is provided a novel image processing apparatus and method for image recognition and classification, the human eye can be relatively faster and more accurately determine whether the unknown object is a known image data from an unknown object.

附图说明 BRIEF DESCRIPTION

[°029]【图1】(A)和⑻是实施例1相关图像处理装置中概念性说明层构造的图。 [° 029] [FIG 1] (A) and ⑻ Examples 1 conceptually related image processing apparatus described layer structure of FIG.

[0030]【图2】㈧〜⑼是对根据各层提取初期候补目标过程举例进行说明的概念图,(E) 是就初期候补目标变换中对象即目标进行概念性说明的图。 [0030] [2] ㈧~⑼ is a conceptual diagram illustrating example according to the initial target candidate extraction process of each layer, (E) is a diagram conceptually illustrating certain i.e. on the initial conversion candidate target objects.

[0031]【图3】对目标的变换和比较判定进行概念说明的图。 [0031] [FIG. 3] and the comparison determination target transform will be explained the concept of FIG.

[0032]【图4】㈧和⑻是就目标重叠情况下的处理举例说明的图。 [0032] [4] (viii) and FIG ⑻ described by way of example on the processing at the target overlap.

[0033]【图5】就图4中目标变换和比较判定进行概念性说明的图。 [0033] [5] FIG comparison target determination and transformation diagram conceptually illustrated on FIG. 4.

[0034]【图6】就使用第2实施例相关图像处理装置的图像处理方法中层的构造进行概念性说明的图。 [0034] [6] FIG configured to use the second image processing method according to the middle of the image processing apparatus related embodiment will be explained conceptually in FIG.

[0035]【图7】图像处理方法中,概念性说明候补层生成处理的流程图。 [0035] The image processing method of [7], the candidate flowchart conceptually illustrating the layer generation process.

[0036]【图8】(A)和⑻是就候补层生成处理中目标的分割进行概念性说明的图。 [0036] [8] (A) and FIG ⑻ is divided conceptually illustrated on the candidate object generation processing layer.

[0037]【图9】㈧〜⑻是概念性说明候补层生成处理中目标的阶层行构造形成构成的图。 [0037] [9] is a conceptual diagram illustrating a configuration ㈧~⑻ candidate hierarchical layer generation process object row structure is formed.

[G038]【图10】(A)和⑻是概念性说明使用第2实施例相关图像处理装置的图像处理方法的变形例的图。 [G038] [10] (A) and ⑻ are conceptually illustrating a modification of an image processing method using the second embodiment of the related image processing apparatus.

[0039]【图11】(A)和⑻是概念性说明层构造的图。 [0039] [11] (A) and ⑻ is a diagram illustrating a conceptual structure of a layer. 【图12】(A)是概念性说明第5实施例相关图像识别装置的构造的框图,⑻是概念性说明图像识别装置中存储装置的构成的框图。 [FIG. 12] (A) is a conceptual block diagram illustrating a configuration of a fifth embodiment related to the image recognition apparatus, ⑻ is a conceptual block diagram showing a configuration of an image recognition apparatus explained in the storage device.

[0041 ]【图13】说明图像识别方法相关第1例的图。 [0041] [13] FIG correlation described image recognition method of the first embodiment of FIG.

[0042]【图14】说明图像识别方法相关第2例的图。 [0042] [14] FIG correlation image recognition method described second example of FIG.

[0043]【图15】(A)〜⑹是说明图像识别方法相关第3例的图。 [0043] [15] (A) ~⑹ illustrates correlation image recognition method of the third example of FIG.

[0044]【图16】(A)是第6实施例相关图像分类装置的处理对象即未知图像的例子,(B)〜 (E)是说明进行分类相关的试纸数据的图。 [0044] [16] (A) is an example of an image that is processed unknown embodiment related image classification apparatus according to a sixth embodiment, (B) ~ (E) is an explanatory diagram relating to the classification of the data strip.

具体实施方式 Detailed ways

[0045]以下结合附图,对本发明的一个较佳实施例作详细说明。 [0045] conjunction with the drawings, a preferred embodiment of the present invention will be described in detail. 但本发明的保护范围不限于下述实施例,即但凡以本发明申请专利范围及说明书内容所作的简单的等效变化与修饰,皆仍属本发明专利涵盖范围之内。 Simple modifications and equivalent scope of the invention but is not limited to the following examples, i.e., whenever in the present invention and the scope of the patent specification the contents of made, are still within the scope of the present invention patent.

[0046] 实施例1: [0046] Example 1:

[0047]本实施例所述图像处理装置,具有CPU、存储装置、显示装置、输入装置和总线等。 [0047] In this example embodiment of the image processing apparatus having a CPU, a storage device, a display device, an input device and a bus or the like. [0048]图像处理装置使CPU等动作,基于已知特定对象物与未知对象物的图像的相似度(确信度),从未知对象物的图像数据,将对应已知特定对象物的图像部分作为目标加以提取。 [0048] Image processing apparatus such as a CPU so that operation of the specific object based on the similarity with known unknown object image (certainty), the image data of the unknown object, the image portion corresponding to the specific object known as target to be extracted.

[0049]此处,所谓目标,是指有一定连贯的图像部分。 [0049] Here, the target means have a coherent image portion. 也就是说,除了从未知对象物图像的未知图像数据提取一部分而得到的图像部分外,连接各图像部分合成的部分也称为对象。 That is, the extracted image portion except a portion of the unknown image data obtained by an unknown object image from the outside, connected to each of the image portion is also referred to as partially synthetic objects. 另外,从未知图像数据中提取的目标中,将可成为单独对应已知特定对象物的目标称为候补目标。 Further, the object extraction from an unknown image data, the target can be a single corresponding to a specific known object is called a target candidate. 不能单独成为候补目标但附带于其他目标,有可能成为新候补目标一部分的目标称为附带目标(即特别目标、非候补目标)。 But alone can not be a candidate target incidental to other targets, may become a new target candidates called incidental part of the target goal (ie special target, target non-candidate).

[0050] 上述图像处理装置中,CPU通过总线,可实现存储装置、显示装置、输入装置相互间的数据收授。 [0050] In the above-described image processing apparatus, the CPU via the bus, the memory device can be achieved, giving or receipt of the display data between devices, each input device. 另外,CPU基于输入装置的操作指示,从存储装置读取制定的程序、数据,执行基于这些程序及数据的各种处理。 Further, the CPU based on the operation indication input means, reading from the storage means to develop a program, data, performs various processes based on these programs and data.

[0051] 具体来讲,CPU将事先读入的图像信息收藏在存储装置中,基于存储装置中的图像信息,进行各种目标的提取、进行各种判定。 [0051] Specifically, the CPU reads the image information in advance in the storage device collection, based on image information storage means, a variety of extraction target, a variety of determination. 图像信息可以是图像装置或其他拍摄装置或者两者结合的拍摄装置所拍摄的图像。 The image information may be a device or another image capturing device or a combination of both imaging devices taken. 比如,摄影装置可以是CCD等固体摄影装置,通过此固体摄影装置检测到的图像可以作为数字图像信号被输出。 For example, such as a CCD imaging device may be a solid-state imaging device, detected by this solid-state image capturing device may be output as a digital image signal.

[0052]存储装置具有程序存储单元和数据存储单元。 [0052] The program storage unit having a storage device and a data storage unit. 程序存储单元具有程序领域,里边存储着多个运转图像处理装置的各种程序等。 Program storage unit having a program area stores various programs and the like inside the operation of the plurality of the image processing apparatus. 数据存储单元有数据领域,里边临时存储着输入指示、输出数据、处理结果等。 Data storage unit has a data field, inside the input instruction is temporarily stored, the output data, processing results and the like.

[0053]显示装置由显示驱动回路、图像显示单元等构成,基于的指令信号,进行必要的显示。 [0053] The display device driving circuit constituted by a display, an image display unit or the like, based on the instruction signal, the necessary display. 显示驱动回路基于从CPU输入的数据,生成驱动信号。 The display driving circuit based on the data inputted from the CPU to generate a drive signal. 基于显示驱动回路输入的驱动信号,图像显示单元进行必要显示。 Based on the display drive signal input circuit, an image display unit for displaying necessary.

[0054]输入装置由键盘等构成,向CPU输出反映操作图像处理装置的操作指令的指令信号。 [0054] The input device includes a keyboard and the like, an instruction signal to the CPU outputs an operation instruction reflects the operation of the image processing apparatus.

[0055]存储装置作为构成要素中的程序存储单元,存储着以下程序:进行处理对象前处理的前处理程序、生成层的层生成程序(比如:第1候补层生成程序、第2候补层生成程序… 第n候补层生成程序,第1附带层生成程序、第2附带层生成程序…第m附带层生成程序等)、 判定与提取对象即已知对象的目标相似度的相似度(确信度)判定程序JP、选择候补对象变换时所利用的目标即增厚•减薄用目标的增厚•减薄用目标选择程序、基于所选择的增厚•减薄用目标,进行目标变换的候补目标增厚•减薄程序。 [0055] The storage device as a component in the program storage unit, stores the following programs: pretreating program processed before processing, layer generation program generation layer (for example: a first candidate layer generation procedure, the second candidate layer generating layer ... n-th candidate program generation program, a first layer included generation program included with the second ... m-th layer generation procedure generating layer included programs), i.e., determines that the object extraction target object is known similarity similarity (degree of certainty ) determination program JP, when the selected candidate target object transformations utilized, i.e. thickening • • thinned with thinning thickened target candidate in the target selection process, based on the selected thinning thickened with • target, the target transform • target thickening thinning program.

[0056]在此,所谓层,是载着根据图像分割所获取的目标的假设的层,根据每个分割模式而生成,最终是含有多个层的。 [0056] Here, the layer is based on the assumption carrying the acquired image segmentation of the target layer, generated in accordance with each of the divided pattern, comprising a plurality of layers eventually. 换而言之,生成由根据整体图像的丨个图像分割模式所得到的多个目标构成的目标群即1个层,因为具有多个图像分割模式,可分别生成与之对应的多个层。 In other words, generating a target group consisting of a plurality of target image segmentation mode according Shu entire image to be obtained, i.e. a layer having a plurality of image segmentation mode as may be generated corresponding to the plurality of layers respectively. 在本实施例中,根据已知手法即各种分割方法,生成多个层。 In the present embodiment, i.e., according to known methods the various segmentation methods, to generate a plurality of layers. 另外,各种层中,所谓候补层,是指含有候补目标的层。 In addition, various layers, a so-called alternate layers, is a layer containing a candidate target. 与之相对,附带层是只仅仅由附带目标(g卩非候补目标、特别目标)构成的层。 In contrast, with only the layer is a layer formed only by the appended target (g Jie non-target candidate, especially target).

[00^7]存储装置作为构成要素中的数据存储单元,存储着以下内容:存储未知对象的图像信息即未知图像数据的未知图像数据存储单元、存储候补层相关数据的第丨〜第n候补层数据存储单元K存储附带层相关数据的第丨〜第!^附带层数据存储单元、存储着候补目标等^关数据的目标数据存储单元(如初期候补目标数据存储单元、暂定候补目标数据存储单元和确定目标数据存储单元)。 [00 ^ 7] storage device as a constituent element in the data storage unit, stores the following: an image information storing unknown objects, i.e. Unknown Unknown image data storage unit of the image data, the first Shu to n-th candidate storage candidate layer related data Shu first layer data storage unit stores K included layers of related data ~! ^ shipped layer data storage unit that stores the candidate target objectives ^ data storage unit related data (e.g., the initial candidate target data storage unit, the provisional target candidate data a storage unit and determines a target data storage unit). 此外,存储装置在数据存储单元中还存储着特定图像数据存储单元。 Further, the storage unit in the data storage means further stores the specific image data storage unit. 特定图像数据存储单元中存储着根据相似度(确信度)判定程序进行判定时所必须的已知对象相关数据。 Specific image data stored in the storage unit when the determination is based on the degree of similarity (degree of certainty) the program determines the necessary data objects is known.

[0058]程序存储单元中,前处理程序是进行图像分析所必要的前阶段的各种处理的程序,比如,对数据存储单元的未知图像数据存储单元中存储的未知图像数据进行各种处理。 [0058] The program storage unit, the first image analysis processing program is a program necessary for various processing of the previous stage, for example, the unknown image data of the unknown image data storage unit for storing data storage unit performs various processes. 另外,第1〜第n候补层生成程序是根据从未知图像数据存储单元读取,在前处理程序中被进行前处理的图像数据,生成各候补层的程序。 The first layer 1 ~ n-th candidate generation program is read from the unknown according to the image data storage unit, image data is subjected to pre-processing of the previous processing program, the candidate program generating layer. 比如,1个候补层生成程序,分割相当于图像整体的未知图像数据的图像,形成分割图像的集合体,即根据1个整体图像,形成多个目标的集合体。 For example, a candidate generator layers, the entire image is divided corresponding to the unknown image data, forming an aggregate image segmentation, i.e., according to an entire image, forming an aggregate of a plurality of targets.

[0059] 在此,各候补层中,将成为候补目标的目标称为初期候补目标。 [0059] Here, each of the layers candidate, will be referred to the initial target candidates candidate target object. 换而言之,第1〜第n候补层生成程序是生成由包含初期候补目标的各种目标构成的候补层的程序。 In other words, the first layer 1 ~ n-th candidate generating program generation program is composed of alternate layers containing various initial target candidate targets. 初期候补目标的决定是基于后期相似度(确信度)判定程序而进行的。 Early decision candidate target is based on the late similarity (certainty) determination procedures carried out.

[0060] 数据存储单元的第1〜第n候补层数据存储单元临时保存着基于第1〜第n候补层生成程序所生成的各候补层相关的数据。 1 ~ n-th layer of the candidate data storage unit [0060] The data storage unit temporarily stores the data relating to each candidate based on the layer 1 ~ n-th candidate generated by the program generation layer. 比如,构成在第1候补层生成程序中生成的候补层的目标的相关信息被存储在第1候补层数据存储单元中。 For example, the target candidate constituting the layer generated in the layer generating a first candidate program related information is stored in the first layer data candidate storage unit.

[0061] 程序存储单元中,第1〜第tn附带层生成程序是分别生成各附带层的程序。 [0061] The program storage unit, a first layer included 1 ~ tn second generation program included in each program are respectively generate layers. 在此,关于不能成为候补目标的附带目标,具体来讲,比如,通过第1〜第n候补层生成程序等,从存储于未知图像数据存储单元的未知图像数据提取目标时的图像处理中,被从构成候补层的各种目标中被排除的目标就是附带目标。 Here, with respect to the target object can not be a candidate, specifically, for example, by a layer of 1 ~ n-th candidate generation program and the like, to extract the unknown image data stored in the image data storage unit is known from the image processing at the target, is excluded from a variety of candidate target layer constituting the target is a target included. 第1候补层生成程序等的图像处理,主要就是通过图像分割进行的,形态各种各样。 Layer generating a first candidate image processing program and the like, is mainly performed by the image segmentation, a wide variety of forms. 在处理中,比如,关于被分割的图像中未满阈值的目标,有时也会一律进行删除处理。 In the process, for example, about the division of the target image is less than the threshold, sometimes harshly deletion process. 也就是说,对于尺寸小的目标,有时也会进行删除处理,因为它不能成为捕捉的已知对象图像的候补目标。 That is, for the small size of the target, and sometimes delete process, because it can not be a candidate target known object image captured. 但是,对于这样的目标,即便单独来看不能称为目的目标,单具有构成目的目标一部分的可能性,有时也可能不完全删除。 However, for this goal, even if the purpose of the individual point of view can not be called a target, it has the potential to constitute a single part of the purpose of the target, and sometimes may not be completely removed. 换而言之,即使是这种被删除的目标,为切实捕捉到目的目标,有时也会使之附带于其他目标,加以应用。 In other words, even if this goal is to be deleted, for the purpose of effectively capture the target, sometimes causing it comes to other objectives, be applied. 本实施例中,如上述,在各种图像处理过程中,也具有这种可能性,即将被从候补中排除的非候补目标作为特别目标(或附带目标)而再次加以采用,即便是像电脑那样进行统一处理的图像处理装置,却可以更加精确提取目的目标。 In this embodiment, as described above, various image processing, also this possibility, about to be excluded from the candidates and non-target candidate to be employed again as a special target (or target included), even as the computer the image processing device as a unified process, but it can be more accurately extracted purpose target.

[0062]数据存储单元的第1〜第m附带层数据存储单元临时存储基于第丨〜第m附带层生成程序而生成的各附带层的数据。 The first m-1 ~ [0062] The data storage unit included with the accompanying data for each layer based on the first to m-th layer Shu generation program included with the generated layer data storage unit temporarily stores.

[0063] 程序存储单元中,相似度(确信度)判定程序是判定基准程序,可数值性判定根据未知图像数据作为候补目标而被提取的目标是否是对应特定对象物。 [0063] The program storage unit, the degree of similarity (degree of certainty) benchmark program determining program is determined, can be determined based on the target value of the unknown image data are extracted as candidates of the target corresponding to whether the specific object. 比如,CPU作为相似度测定单元,从程序存储单元读取相似度(确信度)判定程序,同时,从第1〜第n候补层数据存储单元读取目标,进而,从特定图像数据存储单元读取已知对象物相关数据,基于相似度(确信度)判定程序,通过比较两者的数据,计算确信度数值,进而,CPU作为初期候补目标决定单元根据计算结果,将该目标中相似度高的目标(超过预先规定的阈值的目标)作为初期候补目标。 For example, the CPU as the similarity measurement unit that reads the degree of similarity (degree of certainty) determining program from the program storage unit, at the same time, read from the target candidate 1 ~ n-th layer data storage unit, in turn, is read from a particular image data storage unit take a known object-related data, based on the similarity (degree of certainty) determining program, both the data, and calculates the certainty factor values, Further, the CPU determining unit as an initial target candidate based on the calculation result, the target high similarity target (target exceeds a predetermined threshold value) as the initial candidate target. 关于初期候补目标的彳目息,被存储于初期候补目标数据存储单元。 Head left foot on the initial candidate object information is stored in the data storage unit the initial candidate target. 另外,CPU作为目标判定单兀从存储暂定候补目标的暂定候补目标数据存储单元读取候补目标相关数据, 判定该候补目标有多接近对象物。 Further, the CPU determines that a single candidate Wu read target data from the provisional candidate tentative target storage destination candidate as the destination data storage unit, determines how close the candidate target object. 这些暂定候补目标是通过变换存储初期候补目标的初期候补目标数据存储单元、初期候补目标而得到的。 The tentative initial candidate object is achieved by converting the candidate target data storage unit stores the initial candidate destination, the initial target candidates obtained.

[00M]程序存储单元中,对应候补目标,增厚•减薄用目标选择程序是选择参与判定对象即候补目标的变换•变形的目标即增厚•减薄用目标的程序。 [00M] The program storage unit, corresponding to the target candidate, • thinning thickened with the selection process is to select target objects involved in determining the target i.e. conversion candidate target modification, i.e. thickening • • using the target of the thinning procedure. 另外,候补目标增厚•减薄程序是在进行候补目标变换•变形时,从选择程序中所选择的增厚•减薄用目标中,选择哪些目标用于增厚,哪些目标用于减薄的程序。 Further, the candidate target • thinning thickening candidate program is a certain transformation during deformation •, chosen from the selection process by thinning thickening • targets, to select which target thickening, thinning for which the target program of.

[0065]从初期候补目标数据存储单元中提取的初期候补目标通过候补目标增厚•减薄程序的处理,变换•变形,生成新的目标后,该目标作为暂定候补目标,被临时存储至临时候补目标数据存储单元中。 After the [0065] extracted from the initial data storage unit in the target candidate initial candidate target thickening • thinning processing program, • deformation transform, generating a new candidate target by the target, the target candidate as a provisional target, to be temporarily stored provisional target candidate data storage unit. 暂定候补目标数据存储单元中临时存储的暂定候补目标基于候补目标增厚•减薄程序被变换•变形。 Provisional provisional target candidate target candidate data storage unit temporarily stored thinning thickening • • modification program is converted based on the candidate target. 重复此动作所得到的最终结果目标作为确定目标, 被存储于确定目标数据存储单元。 The end result of this action is repeated to obtain the target as a determination target, the target data is stored in the storage unit is determined. 即确定目标数据存储单元中存储的目标的图像数据是最终应被提取的目标数据。 That determination target object data stored in the storage unit image data is the ultimate object data should be extracted.

[0066]以下,参照图1,就使用具有以上构成的图像处理装置的图像处理方法中层的形成以及初期候补目标的决定,举例进行概念性说明。 [0066] Hereinafter, with reference to FIG. 1, has decided on the use of the initial candidate target formation and image processing method of an image processing apparatus middle above configuration, for example be conceptually described. 图UA)以及i⑻是概念性说明各层的构造的图。 FIG UA) and i⑻ conceptually illustrating the structure of the layers.

[0067]首先,基于上述各层生成程序等,如图HA)所示,许多具有单个或多个目标0B,分别被生成。 [0067] First, based on the program and the like to generate respective layers, as shown in FIG HA), a single or a plurality of targets having a number 0B, respectively, are generated. 也就是说根据分别具有相当于图像分割处理的程序的各层生成程序,分别形成具有多层构造的候补层、附带层。 That layers each having a generating program according to image segmentation program corresponding to the processing, each layer having a multilayer structure formed of the candidate, with layer. 在此,分别生成n层(n彡1)的候补层以及m层的附带层。 In this case, generate the n-layer (n San 1) layer, and a candidate m incidental layer layer. 然后,如图1 (B)所示,从包含于n层候补层中的目标0B中,基于确信度,选择初期候补目标IB(图中阴影部分)。 Then, as shown in FIG 1 (B), the target from the n-layer included in the candidate 0B layer, based on the certainty factor, the initial selection target candidate IB (hatched portion in the figure). 另外,如所述,不从构成m层的附带层的附带目标如中选择初期候补目标。 Further, as described, is not included from the target layer constituting incidental m layer selected as the initial target candidates. 对于所选择的各初期候补目标IB,通过附加其他目标的增厚变换、初期候补目标18减去内部包含的目标的减薄变换,或者既进行增厚变换又进行减薄变换,变化目标,更加精确的提取目的目标。 For each target IB initial candidate selected by the additional thickening transform other objectives, the initial target candidates thinned 18 subtracts the target transform contained within, or perform both thickened and thinned transform conversion, change of the target, more the precise purpose of the extraction target.

[0068] 以下,就使用具有以上构成的图像处理装置的图像处理方法进行说明。 [0068] Hereinafter, an image processing method on the use of an image processing apparatus having the above configuration will be described.

[0069] 首先,CPU作为生成各层的层生成单元,分别读取存储装置的各层生成程序,分别进行层生成处理(参照图1(A)),即第1步骤(步骤S1):各候补层生成处理(提取各候补目标的处理),第2步骤(步骤S2):各附带层生成处理(提取各附带目标的处理)。 [0069] First, the CPU as a generating unit to generate the individual layers, the layers are read to generate a program storage device, layer generation processing (see FIG. 1 (A)), respectively, i.e., a first step (Step S1): Each candidate generation process layer (a process of extracting the candidate target), the second step (step S2): each of the included layers generation process (process target included in each extraction). 更具体来讲,CPU 作为第1层生成单元,基于各层生成程序,生成n层候补层,作为第2层生成单元,基于各附带层生成程序,生成m层的附带层。 More specifically, the CPU 1 as the first layer generating unit generating program based on the layers, the n-layer candidate generating layer, a second layer generating unit, based on the respective layers included generator generating layer included in the m-th layer.

[0070]然后,(PU选择初期候补目标,即第3步骤(步骤S3):初期候补目标选择处理(选择确信度超过阈值的目标)。该第3步骤中,首先,CPU读取被判定的目标,即从数据存储单元的第1〜第n候补层数据存储单元中,读取构成候补层的目标数据(步骤S301),进行该数据的信息分析(步骤S302)。其次,CPU从存储装置读取确信度判定程序的同时,适当读取存储于特定图像数据存储单元中的已知对象物相关数据、确信度阈值等必要数据(步骤S303),基于步骤S3〇2中分析的数据和步骤S:3〇3中读取的数据,判定构成候补层的各目标是否超过确信度阈值(步骤S304)。在步骤S304中,若测定对象目标超过阈值(步骤S304:Yes),CPU将该目标作为初期候补目标存储于初期候补目标数据存储单元中(步骤S305)。另一方面,在步骤S304中,测定对象目标若没有超过阈值(步骤s:3〇4:No),CPU不将该 [0070] Then, (the PU selecting initial candidate target, i.e., the third step (Step S3):. Initial candidate target selection process (selection target sure exceeds a threshold) of the third step, first, the CPU reads the determined target, i.e., from 1 ~ n-th layer data storage unit candidate data storage unit, the read target data (step S301) layer constituting the candidate, information analysis of the data (step S302). Next, the CPU from the memory device and a step of reading data determining program while certainty, proper read specific image data stored in the storage unit related data object is known, it is believed other necessary threshold data (step S303), based on the analysis in step S3〇2 S: 3〇3 read data, determining each of the candidate target layer constituting the certainty factor exceeds a threshold value (step S304) in step S304, the target object if the measured value exceeds the threshold (step S304: Yes)., CPU of the target as the initial target candidates stored in the initial candidate target data storage unit (step S305) on the other hand, in step S304, if the measurement target object does not exceed the threshold value (step s: 3〇4: no)., CPU is not the 目标作为初期候补目标,像处理作为目标变换•变形时可对比的通常目标那样,作为附属层内的一个目标进行存储(步骤S306)。作为初期候补目标决定部的CPU对于构成数据存储单元的第1〜第n候补层数据存储单元的所有目标以上的步骤S301〜步骤S306,决定初期候补目标。 Target as the initial candidate target, the image processing generally target comparable • The modification as the target transformation above, as a target in the subbing layer is stored (step S306). As the CPU candidate target determining unit second initial constituting the data storage unit All the above steps 1 ~ target n-th layer data storage unit candidate S301~ step S306, the target determines the initial candidate.

[0071] 结束第3步骤中上述所述初期候补目标选择处理后,第4步骤:CPU统计所选择初期候补目标的个数并进行编号(步骤S4)。 [0071] After completion of the third step in the above-described initial target candidate selecting process, the fourth step: CPU Statistics The number of the candidate selected and the initial target number (step S4). 在此,N个(N多1)目标是作为第j初期候补目标(Kj <N)被选择的。 Here, N (N is over) as the objective is to select the initial candidate target j (Kj <N) is. 第5步骤:进行初期设定(l—j),即在这N个初期候补目标中,将第1个编号的目标作为最初处理对象即第1初期候补目标(步骤S5)。 Step 5: initial setting (l-j), i.e. the N initial candidate targets, the target of a number of the first object is processed as a first initial candidate object (step S5). 其次,第6步骤:CPU确认对被编号的所有初期候补目标是否处理完毕(是否j>N)(步骤S6)。 Next, Step 6: CPU acknowledges all the initial candidate objects are numbered whether processed (whether j> N) (step S6). 在第6步骤中,若判定还没有完(步骤S6:No),第7步骤:CPU作为候补目标变换单元,从存储装置中读取增厚•减薄用目标选择程序,进行增厚•减薄用目标选择处理(步骤S7)。 In the sixth step, if it is determined yet finished (Step S6: No), Step 7: CPU as candidates for a target conversion means reads from the storage device thickening • thinned in the target selection process, for thickening • Save thin with the target selection process (step S7). 在此,有装入增厚•减薄用目标选择程序, 此程序将作为增厚•减薄用目标而成为处理对象的候补目标内所包含的目标、与该候补目标相邻的目标作为增厚•减薄用目标加以采用。 Here, there was charged with thinning thickening • targeting procedures, this procedure will • thinning thickened with certain targets within the candidate becomes a target to be processed as included, adjacent to the target candidate as a target by • thinning thick to be adopted by the target. 增厚•减薄用目标中,不尽包含构成候补层的目标中作为初期候补目标没有被选择的目标,还包含构成附带层的附带目标。 • thinning thickened with goals, not the goal of containing the target layer constituting the candidate has not been selected as the initial target candidates, also contains incidental targets constitute incidental layer.

[0072] 在第7步骤中,选择增厚•减薄用目标后,第8步骤:CPU从存储装置读取候补目标增厚•减薄程序,提取各种变换模式(步骤S8),比如变换为针对处理对象候补目标,附加(使之附带)增厚用目标,生成新的目标,或者除去减厚用目标,生成新的目标,或者即附加增厚用目标,又减去减厚目标,生成的新目标等多个新候补目标,进行变换处理(步骤S9a), 生成基于各种变化内模式被变换的多个新目标即变换目标,分别测定各变换目标和变换前的目标的确信度(步骤S9b)。 [0072] In the seventh step, by selecting certain thickening • after thinning, Step 8: CPU reads from the storage means a candidate target thickening • thinning procedure, extracts various conversion mode (step S8), such as transformation is a candidate for the target to be processed, additional (so included) thickened with the target, generating a new target, or the target is removed with thickness reduction, to generate a new target, or additionally thickened with certain i.e., the target and subtracting the thickness reduction, Several new candidate target generated new goals, conversion processing (step S9a), the new target is generated based on a plurality of change patterns within the various conversion target converted i.e., the target degree of certainty were measured before and transform each transform target (step S9b). 在此,变换目标和变换前的目标中,有超过事先设定的确信度阈值的目标时,CPU作为目标确定单元,将应该被提取的该目标作为确定目标,作为第j候补目标变换相关信息之一,保存至存储装置的一个领域内(步骤S10)。 When the target before this, conversion targets, and transformation, target certainty threshold value exceeds a pre-set, the CPU as the target determination unit, the target to be extracted as the determination target, as the j-th candidate target conversion infos one of a field stored in the internal storage device (step S10). 其次,CPU作为目标确定单元,对步骤S8〜S10中变换后被保存的新候补目标即变换目标,比较基于确信度判定程序的确信度,特定确信度最高的目标(步骤S11)。 Next, the CPU determines as the target unit, the candidate for a new target after step S8~S10 saved transform or transfer target, based on comparison certainty certainty factor determination procedure, the highest target certain degree of certainty (step S11). 在此,将新候补目标即变换目标中确信对最高的目标称为最佳候补目标。 Here, the new candidate for the highest goals confident target as the best candidate target or transfer targets. 其次,CPU确认步骤S11中被特定的最佳候补目标的确信度是否比变换前的候补目标的确信度上升(步骤S12)。 Next, CPU is confirmed in step S11, a particular degree of certainty whether the best candidate target increased (step S12) a candidate target certainty than before conversion. 在步骤S12中,如果确认确信度上升,即判断变换后的候补目标中确信度最高的最佳候补目标暂时是最高确信度的目标时(步骤S12: Yes),取代变换前的候补目标,将最佳候补目标作为新的处理对象即暂定候补目标。 In step S12, if it is confirmed certainty increased, i.e., the best candidate is determined highest target candidate target the transformed certainty temporarily highest certainty factor is the target (Step S12: Yes), the candidate target before conversion substituted, the the best candidate as a new treatment target objects candidate ie provisional target. 改写暂定候补目标数据存储单元中候补目标信息(步骤S13),将该暂定候补目标作为处理对象,在此重复步骤S9〜S12的动作。 Rewriting the provisional target candidate data storage unit candidate object information (step S13), the provisional target candidate as the process target, at step S9~S12 repeat operation. 另一方面,在步骤S12中,确认确信度没有上升,即判断变换前的候补目标是具有更高确信度的目标时(步骤S12:No),判定变换前的候补目标是最佳目标,若该候补目标超过确信阈值,将其作为确定目标之一进行处理,同时结束第j候补目标变换处理(步骤S14),移向(步骤S15)下一个初期候补目标的处理(j+1—j),同样进行上述处理,直至所有初期候补目标的处理结束(步骤S16: Yes)。 On the other hand, in step S12, it is confirmed not increased certainty that the target is determined before conversion candidate having a higher degree of certainty the target (Step S12: No), the candidate targets determined by the optimum target before conversion, if the candidate target sure exceeds a threshold value, which is treated as one determination target, while the end of the j-th candidate target conversion processing (step S14), and moves (step S15) a candidate initial processing target (j + 1-j) Similarly the above process until all of the initial candidate target process is completed (step S16: Yes).

[0073]以上,通过图像处理装置,对所有N个(N多1)初期候补目标进行变换或变形,结束确定目标的提取。 [0073] or more, for all N (N is over) the initial target candidates for modification or conversion by the image processing apparatus, extracting a determination target ends. 此时,步骤S10中被保存提取对象即确定目标在1种类的候补目标变换中, 可能生出多个。 In this case, step S10 is stored in the object extraction target i.e. to determine the type 1 candidate target transformation, may give birth to a plurality. 此时,比如,可通过其他判定方法、或从人眼确定的多个确定目标中最终选择1个确定目标。 In this case, for example, it may be determined by other methods, or the human eye from the determined plurality of targets is determined in a determination target finally selected. 通过在1个候补目标变换中,选择多个具有可能性的确定目标,可降低应被提取的目标的遗漏的可能性。 By a candidate target transform, select a target having a plurality of possibilities to determine the possibility of missing a target to be extracted can be reduced.

[0074]图2¾图3是就上述图像处理装置的图像处理中的变换候补目标过程进行更具体说明的模式不意图。 [0074] FIG 2¾ FIG. 3 is a more detailed description of the mode conversion candidate on the target image processing procedure of the image processing apparatus is not intended. 比如,将具有如图2 (A)所示长方形范围的整体图像肋作为未知图像,进行图像处此时,重叠阶层性设置的分割图像的层(比如,参照图丨(A))中各领域时的图如图2®)所示。 For example, having a in FIG. 2 (A) shown in a rectangular range of the entire image of the ribs as an unknown image, for this case, the divided image superimposed hierarchical layer disposed at the image (for example, see FIG Shu (A)) of each field FIG 2® Very view) FIG. 整体图像HD中的领域即目标〇B的一部分如图2 (C)所示,被选择作为初期候补目标IB。 That part of the target 〇B entire image field HD in FIG. 2 (C), the candidate is selected as the initial target IB. 进而,如图2⑼所不,对于N个初期候补目标IB,为了按顺序进行处理,进行了编号。 Further, as shown in FIG 2⑼ are not, for the IB of the N initial target candidates, to be processed sequentially numbered. 图2⑻显示的就是针对N个中1个初期候补目标18 (图中显示的是第丨个初期候补目标IB1^ 而选择的增厚•减薄用目标的一个例子。在图例中,丨点短划线所示第丨目标是包含于第丄初期候补目标IB1中的目标。换而言之,第丨目标是将第〖初期候补目标IB1作为父目标的子目标C0。虚线所示第2目标N0是与第1初期候补目标IB1邻接的相邻目标船。属于各候补层的目标中,除了作为初期候补目标IB未被采用的目标外,属于各附带层的附带目标(特别目标) 也可被采用作为这些目标(比如第1目标、相邻目标)。也就是说,候补目标变换时,通过将这些不采用目标作为增厚•减薄用目标,虽作为初期候补目标IB未被采用,但确定可以作为目的目标最终被提取时,可以将其作为目标一部分再次被米用。更具体来讲,如图3所不,作为1个变换模式(变换模式A),附加与第1初期 FIG 2⑻ for N is displayed in an initial candidate target 18 (shown in FIG Shu is a first initial target candidates selected IB1 ^ • One example thickened thin target. In the illustrated example, the short points Shu the first objective Shu target comprising a target candidate in the first Shang initial IB1. in other words, the goal is to first 〖Shu initial candidate parent object as the target IB1 subgoal C0. as shown in dashed chain line in the second target N0 is the initial candidate target IB1 adjacent to the target vessel adjacent the first target candidate belonging to each layer in addition to the target candidate as an initial target IB is not used, are included in each layer is included with the target (specific targets) may be is adopted as the target (such as a first target, adjacent to the target). In other words, when the candidate target transformed by these goals is not used as a thickening • thinning with a goal, although as the initial target candidates are not using IB, but the final determination may be extracted as a target object, it can be used as part of the target meter is again used. more specifically, FIG. 3 does not, as a mode conversion (transform mode A), with the first initial additional 补目标IB1邻接的相邻目标NO中的1个,生成第1新的目标A1 (图中阴影部分)。另外,作为别的变化内模式(变换模式B),减去第1初期候补目标IB1中内含的子目标C0中的1个,生成第2新的目标Bl(图中阴影部分)。其他的,比如, 进一步作为其他变换模式(变换模式C),附加于第1初期候补目标IB1邻接的相邻目标NO的同时,减去内含的子目标C0,生成第3新的目标Cl(图中阴影部分)。如此,叠加种种(有限的) 变换•变形模式,对此分别进行相似度测定,如双箭头所示,通过比较这些变换候补目标, 可决定最佳候补目标。进而,通过比较这些目标与第1初期候补目标IB1,可判断变换前后确信度是否有上升。对于变换后的候补目标(即暂定候补目标),根据需要,重复上述相同动作,最终可取的确定目标。 IB1 adjacent target complement adjacent target in a NO, generates a first new target A1 (hatched portion in the figure). Further, as another change in the mode (B mode conversion), subtracts the first initial candidate target IB1 C0 contained in the sub-target in a, to generate a second new target Bl (hatched portion in the figure). others, for example, as a further other mode conversion (transform mode C), in addition to the first initial candidate target IB1 NO contiguous adjacent target while subtracting subgoal contains C0, and generates a third new target CI (hatched portion in the figure). thus, the various superimposed (limited) • deformation mode conversion, which were similar Determination, as indicated by the double arrows, these conversion candidates by comparing the target, the target may determine the best candidate. Furthermore, these objectives by comparing the first candidate to the initial target IB1, the certainty factor may be determined whether there is increased before and after transformation. for the transformed candidate target (i.e., the candidate tentative target), if necessary, repeat the same operation, the final determination of desirable goals.

[0075]以下,参照图4等,就上述变换模式的一个变形例进行说明。 [0075] Hereinafter, with reference to FIG. 4 and the like, will be described a modification of the conversion mode. 如图4(A)中2点短划线所示,对于初期候补目标,有时也会重叠其他目标。 , The candidate for the initial target, sometimes overlap other objects in FIG. 4 (A) two-dot dashed lines. 存在这样的重叠目标SS时,决定变换模式时,可考虑利用重叠目标SS。 When there is a certain overlap SS, determining transform mode contemplated using overlapping target SS. 比如,如图4⑻所示,将包含元候补目标CC(图中显示为第1 初期候补目标IB1)与重叠目标SS的目标作为父目标PP,通过第1候补目标CC的轮廓与重叠目标SS的轮廓,划定5个子目标C0。 For example, as shown in FIG 4⑻, including candidate target element CC (shown as a first candidate initial target IB1 drawing) overlaps with the target as the target parent object SS PP, CC by the first candidate target contour overlaps with the target SS contour, delineation of five sub-goals C0. 如图5所示,基于通过组合子目标C0的有限变换模式,可生成第1、2、3新目标D1、D2、D3等,对此,进行上述同样的基于确信度的比较。 5, based on the combination of sub-finite transform mode C0, a target, the new target may be generated 1,2,3 D1, D2, D3, etc., which, for the same degree of certainty based on the comparison. 比如,关于尺寸,规定下限阈值,对于子目标C0中尺寸在该下限阈值以下的目标,可作为附带目标予以处置。 For example, with regard to size, the lower limit threshold value for the sub-goals C0 in size in the lower threshold of the target, the target may be disposed of as incidental. 即设置包含该子目标的附带层,使其属于附带层,生成这样的构成。 That layer is provided containing the incidental subgoals, it belongs to the layer incidental to generate such a configuration.

[0076]如上,本实施例相关图像处理装置以及使用该装置的图像处理方法中,CPU作为候补目标变换单元,变换初期候补目标,生成新的候补目标。 [0076] As above, the present embodiment related to an image processing apparatus and an image processing method using the apparatus, the CPU as a candidate target transform unit converts the initial target candidates, generating a new candidate targets. 即,对从拍摄的未知对象物写真等未知图像数据提取的目标进行必要的变换,可提取对应已知对象物的新目标。 That is, the extracted target captured images and other unknown unknown object image data necessary conversion, the new target may be extracted corresponding to the known object. 特别是,上述中,CPU作为相似度测定单元,读取确信度判定程序,基于测定的相似度,进行成为最初候补的初期候补目标的提取,进而,CPU作为候补目标变换单元,读取增厚•减薄用目标选择程序、候补目标增厚•减薄程序,变换初期候补目标。 In particular, the above, the CPU as the similarity measurement unit that reads certainty determination program based on the measurement of the degree of similarity, a target for a candidate to become the initial extraction of the first candidate, in turn, transform the CPU as a candidate target cell, read thickening • thinning with target selection procedures, candidates target thickening • thinning program, transforming the initial target candidates. 进而,CHJ对于变换的目标,进行相似度测定,根据需要,重复变换动作,作为目标确定单元,确定应该提取的目标。 Further, the target CHJ transformation, similarity measured, if necessary, repeat the conversion operation, a target determination unit, determines the target to be extracted. 此时,关于从所拍摄未知对象物写真等未知图像数据所提取的目标,进行必要的变换,可提取对应已知对象物的新目标。 At this time, from the target object images and other unknown unknown extracted image data of the captured, the necessary conversion, the new target may be extracted corresponding to the known object.

[0077]如上所述,关于附带层或附带目标的生成,对于利用根据第1〜第n候补层生成程序生成候补层时被排除的目标时,进行了举例说明,但并不仅限于此,还可应用其他各种方法。 [0077] As described above, certain incidental or incidental generation layer, for use when a target n-th candidate 1 ~ According layer generating program is excluded candidate generating layer, are illustrated by, but is not limited thereto, but also other various methods can be applied.

[0078]另外,如上所述,在步骤S10中,基于事先规定的确信度,进行决定应该提取的确定目标的判断,不仅限于确信度的阈值,比如,基于轮廓层,也可提取确定目标。 [0078] As described above, in step S10, based on a predetermined degree of certainty beforehand, determination is a determination target is determined to be extracted is not limited to the threshold degree of certainty, for example, based on the profile layer, the target may be determined to extract. 在此,所谓轮廓层,比如,是指由轮廓相关信息组成的层,该轮廓是指划定图像分割处理中成为基准的分割领域的轮廓。 Here, the profile layer, for example, refers to a layer composed of the contour information, the contour is delineated image division processing means becomes a reference field contour segmentation. 这样的轮廓可以显示的是成为应提取对象即目标的外沿。 Such a profile can be extracted to be displayed is the object that is the outer edge of the target. 这样的轮廓相关信息有时也可以直接显示目标的范围。 Such profile information may be displayed directly target range. 比如,成为判断对象的目标的外沿与该轮廓的全部或一部分一致时,可判断其为应提取的确定目标。 For example, the outer edge of the target object is determined to become an object judgment coincides with all or a portion of the contour, it can be determined which is to be extracted. 此时,关于目标的提取,基于轮廓信息,可切实进行所需目标的提取。 At this time, the extraction of the target, based on the profile information, can effectively extract the desired objectives.

[0079] 另外,上述中,基于确信度判定程序进行的相似度测定结果,进行初期候补目标的提取,但关于初期候补目标的提取,比如,也可以设定目标尺寸、颜色等各种数据阈值,基于该阈值进行提取。 [0079] Further, in the above determination program based on the certainty factor for similarity measurement result, the initial candidate object extraction, the extraction on the initial candidate object, for example, may be set target size, color and other data threshold , based on the extracted threshold value. 也就是说,并非通过使用相似度概念,根据其他数据值,决定初期候补目标,进而,不使用相似度概念(比如,利用轮廓层),变换该初期候补目标,可提取新的目标。 That is, not by using the concept of similarity, based on other data value, determines the initial target candidates, and further, does not use the concept of similarity (for example, using a profile layer), transforming the initial candidate target, the new target can be extracted.

[0080] 另外,在其他处理中也可利用上述各处理中所得信息。 [0080] Further, other processing may also utilize the information obtained in the above-described process. 比如,在重复阈值进行计算的处理中,即使直到最后也没有超过阈值,也可将该处理中对象目标相关信息作为图像分割处理等相关信息加以利用。 For example, the process is repeated to calculate the threshold value, even until the last did not exceed the threshold, the process may be subject target information is utilized as the image division processing information and the like.

[0081] 实施例2: [0081] Example 2:

[0082] 以下,通过图6,就使用本发明第2实施例相关图像处理装置的图像处理方法进行说明。 [0082] Here, Fig. 6, on the use of the image processing method of the second example of related image processing apparatus of the embodiment of the present invention will be described. 本实施例所述图像处理装置是第1实施例的变形例,关于图像处理装置的构造,与第1 实施例的图像处理装置相同,在此省略说明。 This embodiment of the image processing apparatus is a modification of the first embodiment, regarding the configuration of an image processing apparatus, the image processing apparatus of the first embodiment are the same, description thereof is omitted here. 关于图像处理顺序,除去特别说明的情况,与实施例1的流程相同。 The image processing sequence, removed particularly stated, the same procedure in Example 1.

[0083]图6就本实施例下的图像处理方法中层的构造进行概念性说明。 [0083] FIG. 6 method of construction at the middle of an image processing according to the present embodiment will be explained conceptually. 换而言之,作为为了通过增厚•减薄变换等进行目标提取所进行的各种处理的前段处理,图6显示的是各层的构成被确定的一种状态。 In other words, as various processes for converting thinning thickening • like object extraction performed by the pre-stage processing, FIG. 6 shows a state of the respective layers constituting determined.

[0084]本实施例中,多个层中包含了上方层和下方层这样阶层性构造的层群。 [0084] In this embodiment, the plurality of layers comprises a layer group and over the layer below this layer hierarchical configuration. 更具体来讲,构成上方层的目标包含着构成下方层的目标。 More specifically, the target constituting the upper layer contains the target constituting the lower layer. 反而言之,构成下方层的目标被包含于构成上方层的目标中。 Conversely, the lower layer constituting the target is included in the target constituting the upper layer. 也就是说,上方层的目标是父目标,下方层的目标是其子目标,上方层和下方层形成了父子构造。 That is, above the target layer is a parent object, which is below the target sub-target layer, the upper layer and the lower layer is formed of a parent-child structure. 特别是如图6所示,关于所有的候补层,形成阶层性的构造。 As shown in FIG 6 in particular, with respect to all the candidate layer, the hierarchical structure is formed. 比如,构成第P候补层的所有目标是构成相对处于下方层即第P+1候补层所有目标的父目标, 进而,构成第P+1候补层的所有目标是构成相对处于下方层即第P+2候补层的所有目标的父目标,形成这样的关系。 For example, all targets constituting the P candidate layer constituting the parent object relatively in the lower layer i.e., the first P + 1 candidate layer all targets, and further, constituting all target first P + 1 candidate layer constituting relatively in the underlying layer ie P All the parent target candidate targets +2 layer, forming such a relationship. 如此,形成父辈下面有子辈,进而下面有孙辈这样的阶层性亲子构造时,下方侧的目标全部被包含于上方侧的目标中。 Thus, when forming the following sub generation parents, parent-child hierarchical below further configured such grandchildren, the target lower side are all included in the upper side of a target. 比如,不需要生成第1实施例的图4等所示重叠部分,不需要再构造每个群,可进行增厚•减薄变换。 For example, no need to generate overlapping portion shown in FIG. 4 of the first embodiment and the like, do not need to re-configuration of each group can be converted thickening • thinning. 换种角度,也可以事先进行图4 等所示重叠部分处理,再生成具有父子关系的构造。 Put another angle, the overlapping portion may be performed in advance the processing shown in FIG. 4 and the like, and then generates a configuration having a parent-child relationship.

[0085]作为形成上述父子构造的方法之一,比如,在候补层生成程序的选择中,可考虑事先选择基于进行满足上述父子关系条件的图像分割的程序,如图4等例子所示,为形成上述父子构造,关于既存候补层生成,可再变成。 [0085] As one method for forming the parent-child structure, for example, in the selection candidate program generating layer, may be considered in advance based on the above selection procedures paternity image segmentation condition is satisfied, the example shown in FIG. 4 and the like, is forming the parent-child structure, existing on the candidate generating layer may be further becomes.

[0086]以下,参照图7所示流程图,就作为一个为了根据己存候补层生成具有上述阶层构造的候补层,而再编成候补层的图像处理的例子,进行目标分割、再分割等,满足父子关系条件,且构成在规定范围内捕捉各目标尺寸的情况进行了说明。 [0086] Hereinafter, with reference to the flowchart shown in FIG. 7, in order to generate a candidate as a layer having the above configuration according to the hierarchical layer has stored candidate, while reorganizing example of an image processing layer candidate, the target division, subdivision, etc. , parent-child relationship satisfies the condition, and the composition of each of the captured target size within a predetermined range has been described.

[0087]首先,CPU读取关于既存候补层内目标的信息,测定各目标的尺寸(步骤S201),测定尺寸是否超过事先确定的上限阈值(步骤S202)。 [0087] First, the CPU reads information about the target within an existing layer candidate, measuring the size of each object (step S201), determination whether the size exceeds the upper threshold determined in advance (step S202). 在步骤S202中,检测到上限阈值以上尺寸目标时(步骤S2〇2:Yes),CPU提取该目标(步骤S2〇3),进行分割处理(步骤S204),直至所有目标尺寸部在上限阈值以下,重复分割处理动作(步骤S205)。 In step S202, the detected size of the target upper limit threshold or more (Step S2〇2: Yes), CPU extracts the target (step S2〇3), dividing processing (step S204), until all the target sizes portion below the upper threshold repeated segmentation process operation (step S205). 图8是显示以上分割处理的模式不意图。 FIG 8 is a more segmentation mode treatment is not intended. 即图8 (A)中,属于第q候补层的目标M〇的尺寸为阈值以上时,根据事先规定的分割方法,分割此目标M0,细分为图8⑻所示多个分割目标M〇d。 I.e., in FIG. 8 (A), belonging to the first layer q candidate target M〇 size is less than the threshold value, according to a predetermined division method in advance, dividing this goal M0, the target segment is divided into a plurality of FIG 8⑻ M〇d . 比如考虑与对应已知对象物图像数据尺寸的比较结果,若有必要,进而对这些分割目标Mod进行再次分割。 Consider, for example corresponding to the known size of the target image data of a result of comparison, if necessary, further segmentation of these objectives is divided again Mod. 分割前的目标M0与分割后的分割目标Mod具有父子构造关系。 M0 goal before the division has a parent-child relationship between structure and segmentation target Mod divided. 但是,比如,关于尺寸,规定了下限阈值,分割目标Mod中,对于尺寸在该下限阈值以下的更小目标,可以作为附带目标进行处理。 However, for example, on the size, a predetermined lower threshold, the target segmentation Mod, with respect to the size of the lower threshold value is smaller targets, the target can be treated as an incidental. [0088]返回图7,在步骤S2〇2中,没有检测出阈值以上尺寸目标时(步骤S202:N〇),或者通过步骤S203〜步骤S2〇5的动作,所有目标变为阈值以下尺寸时(步骤S2〇5: Yes),保存这些目标(步骤S206)。 When: (N〇 step S202), or by the action of step S203~ S2〇5 step, all of the target becomes a threshold value when the size [0088] Returning to Figure 7, in step S2〇2 not detected above a threshold size of the target (step S2〇5: Yes), save these goals (step S206). 其次,CPU从被保存的目标中,提取(步骤S2〇7)没有子的目标(最小目标)。 Next, the CPU is saved from the target, the target is no child (minimum target) extracting (step S2〇7). 即以上下关系看父子构造时,提取位于最下层的目标。 That is, when the above configuration Relations Sons, lowermost extraction target. 其次,CPU形成具有父子构造的层群的条件是否满足,即对于所有没有子的目标,分别确认是否有父目标(步骤5208),在步骤S2〇8中,若判定没有满足条件(没有父目标)(步骤S208: No)后,统和该无子的目标所属层与统一层内的目标,生成成为父目标的目标(步骤S2〇9)。 Next, the CPU having a layer formation condition of the group meets a parent-child structure, i.e. for all sub-goals are not respectively confirms whether the parent object (step 5208), in step S2〇8, if the determination condition is not satisfied (no parent object ) (step S208: no), the target object belongs in a unified layer of the layer system and the non-promoter, to generate a target parent object (step S2〇9). 根据该新生成的父目标,生成构成中上方层(步骤S201)。 According to the newly generated parent object, the upper layer (step S201) configured to generate. 重复上述生成上方层的处理,直至目的条件满足,即最大目标不再是阈值以上尺寸(步骤S208:Yes),以此形成再编成的新候补层(步骤S211)。 The generation process is repeated over the layer until the object condition is satisfied, i.e., the maximum target size is no longer than a threshold value (step S208: Yes), thereby forming a new layer candidate (step S211) and then compiled.

[0089]通过如上动作,根据属于下方层的目标生成上方层的处理。 [0089] With the above operation, the upper layer is generated based on the processing target belongs to the lower layer. 图9就是模式性示意此处理的图。 FIG 9 is a schematic of this process schematically in FIG. 为了形成如图9 (A)所示阶层性构造,首先,如图9 (B)所示,集中位于没有子目标的最下方层的目标,生成最下方层。 To form the hierarchical structure shown in FIG. 9 (A), first, as shown in FIG 9 (B), the target is centrally located in the lowermost layer is not subgoal generated lowermost layer. 在此,将最下方层称为第丨层。 Here, the bottom layer is referred to as Shu layer. 然后,如图9(〇所示,进行统和处理构成最下方层的目标,形成其父目标,生成集合了该父目标的层。在此,新生成的层称为第2层。以下,同样,如图9 0:)所示,通过统和处理构成第2层的目标,形成其父目标, 如图9⑼及9⑻所示,生成新层(第3层)。 Then, as shown in FIG 9 (shown square, and the system for processing object constituting the lowermost layer, is formed parent target, generating a set of layers of the parent object. Here, newly generated layer is referred to below the second layer, Similarly, as shown in FIG. 90 :), the second layer constituting the target by the system and processed to form a parent target, as shown in FIG. 9⑼ 9⑻ and generate a new layer (layer 3). 如上述步骤S208—样,比如,就目标尺寸等设定阈值,一直重复以上动作直至超过该阈值。 The above steps S208- like, for example, setting a threshold value to the target size and the like, the above operation is repeated until the threshold is exceeded. 可以形成由一定范围内尺寸的目标构成的父子构造的候补层。 Sons alternate layers can be formed within a certain range by the size of the target constituting the structure. 关于目标统和方法,可有各种规定,比如,关于统和目标的处理,在同一层内, 可统和最相近的目标。 About the target system and method, various provisions may be, for example, on the processing system and the target, in the same layer, and may be the closest target system. 或者,若已经具有父子构造时,可优先其父子关系。 Alternatively, if the parent-child structure has, its priority parent-child relationship. 换而言之,同一层内,对于原本就存在共通的父目标,即同一层内属于兄弟关系的目标,优先进行统和。 In other words, within the same layer, for there is already a common parent object that belongs to the target of brotherhood within the same layer, and a priority system. 避免发生如图4所示的重叠目标,同时要保留存在的父子构造。 Avoid overlapping target occurs as shown in FIG. 4, at the same time to retain the existing configuration Sons. 另外,容许父子构造复杂交错的状态。 Furthermore, complex structure allowing Sons staggered state.

[0090]另外,以上例子时,作为层再编成的前奏,在步骤S201〜步骤S204中,对阈值以上尺寸目标,强制进行分割。 [0090] Further, when the above example, as a prelude to re-woven layer, in step S201~ step S204, the size of the target above the threshold force is divided. 以此,可以避免应被提取的目标被埋在大块目标中而被遗漏的情况发生。 This can be avoided should be extracted object buried in chunks objectives and circumstances of the occurrence are missing. 关于此处的分割,可采用各种方法,比如,格子状统一分割等将目标尺寸缩小的手法等。 Divided on herein, various methods, such as, uniform grid pattern of the target segmentation technique downsizing and the like.

[0091]图10所示意的就是在具有上述阶层性构造(父子构造)的层群的情况下提取目标的例子。 [0091] Examples of the FIG. 10 is a schematic of an extraction target in the case where the layer having the above hierarchical cluster configuration (parent-child structure). 图10(A)所示的是从最下方层第P+2候补层安顺序进行变换处理的示意图。 FIG 10 (A) is a schematic view of the transformation process from the candidate sequence Layer Security lowermost layer P + 2 as shown. 具体来讲,就是从属于最下方层的候补层中,将相似度高的目标作为初期候补目标加以选择,进行变换的提取方法。 Specifically, the candidate is dependent on the lowermost layer of the layer, the high similarity be selected as an initial target a target candidate, transformation extraction method. 此时,比如,在最初阶段,同一层内,仅仅进行增厚处理,在同一层内若相似度未达到既定阈值,对将初期候补目标作为子目标包含在内的最近上方层父目标再次进行增厚即减薄处理,一直重复此动作直至相似度达到既定阈值,当相似度达到目的阈值时, 提取目标作为确定目标。 In this case, for example, in the initial stage, the same layer, only thickening process performed in the same layer when the similarity does not reach a predetermined threshold value, the target of the initial candidate as the most recent layer over the parent goal subgoals inclusive again That thinning thickening process repeats this action until the similarity reaches a predetermined threshold value, when the object reaches a threshold similarity, extracting a target as a determination target. 另一方面,图10 (B)示意的是从最上方层即第1候补层开始按顺序进行变换处理的例子。 On the other hand, illustrated in FIG. 10 (B) is an example of that is the first candidate in sequence layer starts from the conversion processing uppermost layer. 即,将最上方层中相似度高的目标作为初期候补目标加以选择,对该初期候补目标最近的下方层中构成初期候补目标的子目标,在最初阶段仅进行减薄处理, 之后通过增厚、减薄,根据需要进行重复,直至提取达成目的相似度的目标,并将其作为确定目标。 That is, the uppermost layer having high similarity as an initial target to be a candidate target selection, the initial candidate child object constituting the initial candidate target object nearest the lower layer, only the thinning process performed at the initial stage, after passing through the thickening , thinning, repeated as necessary, until the extracted target object to achieve the degree of similarity, and as a determination target. 也可以同时进行图10(A)和图10⑻的动作,提取共通目标时,将该目标作为确定目标。 It may be performed simultaneously (A) and FIG 10⑻ operation of FIG. 10, the common extraction target, and the target as a determination target.

[0092] 实施例3: [0092] Example 3:

[0093]本实施例相关图像处理装置是第1实施例等的变形例,就图像处理装置的构造来讲,与第1实施例的图像处理装置相同,在此省略图示及说明。 [0093] The present embodiment is an embodiment related to an image processing apparatus of the first embodiment such modified embodiment, it is configured in terms of the image processing apparatus, the image processing apparatus of the first embodiment is the same in this illustration and description thereof will be omitted.

[0094]本实施例中,生成不仅仅是由候补目标、附带目标,还有其他候补目标构成的其他候补层。 [0094] In this embodiment, not only generated by the target candidate, with the goal, there are other candidate target layer composed of another candidate. 就这一点来讲,与上述实施例不同。 In this point of view, the above-described embodiment. 在此,所谓其他候补目标,是指对应虽然是已知对象物,单并非是提取对象的对象物的候补目标。 Here, the target of another candidate, although the corresponding means known object, the object extracting a single candidate is not a target object. 换而言之,其他候补目标与候补目标不同,是不能成为目的未知对象物的目标。 In other words, candidates and other candidates target different goals, it can not be an end in unknown object target. 关于其他候补目标,是通过目标提取处理、目标变换处理所不能得到的目标。 The other target candidate, extracted by the processing target, the target can not be a target of the conversion process. 也就是说,关于其他候补目标,不能成为候补目标,也不能称为附带目标,是从处理对象中被排除的。 That is to say, with regard to other candidates goal, the goal is not to become a candidate, can not be called incidental goal, from the processing object to be excluded.

[0095] CPU作为第1层生成单元,基于各层生成程序生成n层候补层。 [0095] CPU generating means as the first layer, n layer, the respective layers based on the candidate generating program generating layer. 作为第2层生成单元, 基于附带层生成程序,生成m层附带层。 Generating means as the second layer, based on the supplementary layer generation procedure generating layer comes m layer. 同时,进而作为第3层生成单元,基于其他候补层生成程序,如图11(A)所示,生成k层其他候补层。 Meanwhile, as the third layer and further generating means generating layer based on other candidate program, as shown in FIG 11 (A), the layer generating k candidate other layers. 但是,这些k层其他候补层被排除在外。 However, these layers of other candidate k layers are excluded. 也就是说,是与层形成以后的图像处理无关的。 That is, after forming a layer independent of image processing. 具体来讲,CPU提取候补层即附带层做候补目标、 附带目标的同时,也提取其他候补目标,之后,对不能成为对象的其他候补层,做排除处理(参照图11⑻)。 Specifically, the CPU candidate extraction layer that is included with layers do candidate target while the target is included, but also other extraction target candidate, then, not be subject to other candidate layers do elimination processing (refer to FIG 11⑻). 关于其他处理步骤与实施例1相同,在此省略说明。 The other process steps are the same as in Example 1, description thereof will be omitted here.

[0096]程序存储单元中,其他候补层生成程序通过从整体图像提取判定对象之外的特定对象物相关目标,生成层。 [0096] The program storage unit, the other layer generated candidate program by the specific object is determined from the relevant target objects other than the entire image extraction, generation layer. 如上所述,此处所生成的目标即其他候补目标、由其他候补目标构成的其他候补层等相关信息在之后的处理中被排除在外。 As described above, i.e., the generated target other candidate targets herein, other candidate other layer and the like made of certain information in the candidate after processing are excluded. 根据应该提取的对象的属性、 层的正确性等,如本实施例那样,通过事先去除不能成为对象的其他候补目标,可更加快速进行后阶段的处理即增厚•减薄处理。 The properties of the object to be extracted, like the correctness layer, as in this example embodiment, by previously removing the other can not be a candidate of the target object can be processed more quickly, i.e., after thickening stage • thinning process.

[0097]根据其他候补层生成程序,生成多个其他候补层的图像处理中,被排除在外的目标,可将其作为附带目标加以利用。 [0097] According to other layers candidate generating program, image processing to generate a plurality of layers of other candidates, the target excluded, it can be used as the target comes.

[0098] 实施例4: [0098] Example 4:

[00"]以下,就本发明第4实施例相关图像处理装置的图像处理方法进行说明。本实施例相关图像处理装置是第1实施例等的变形例,就图像处理装置的构造,因与第丨实施例的图像处理装置相同,在此省略图示及说明。另外,关于图像处理顺序,除去特别需要说明的情况外,与实施例1相同。本实施例中,想要完全分类为多个种类已知对象物时,务必在某个步骤中利用除外对象即其他候补层生成单元,则不会存在事先被排除(除外)的目标。 [00 "] Hereinafter, on the image processing method related image processing apparatus according to a fourth embodiment of the present invention. The present embodiment related to an image processing apparatus is a first embodiment of such modification, is configured to image processing apparatus, because the the image processing apparatus of the same embodiment of Shu embodiment, illustration and description thereof will be omitted herein. Further, the image processing sequence, except for special cases Incidentally, same as in Example 1. in this embodiment, a multi want to completely classified when the object is a known type, except for using an object in a sure step, i.e. other layers candidate generating unit is not to be excluded the presence of the target in advance (excluded).

[0100] 上述实施例中,将已知对象物作为1种类进行了说明,但也会存在从1个整体图像想要提取2个种类以上的已知对象物的情况。 [0100] In the embodiment described above, the known object has been described as one kind, but also from the presence of an entire image to be extracted is known where the object two or more species. 比如,关于1个整体图像,有时会想就多个被反应的目标,将其完全分类为多个种类的已知对象物。 For example, on an overall image, sometimes I think it is more objective response will be classified as more fully known object types. 为此,本实施例中,就进行多个种类图像提取的图像处理装置及方法进行说明。 For this reason, in this embodiment, to the image processing apparatus and method for extracting a plurality of types of images will be described. 换而言之,相对于第3实施例中所示其他候补层的其他候补目标,仅对于候补层进行的目标提取处理相当于个别增厚•减薄等的目标提取的各种处理。 In other words, as shown with respect to other candidate target layer, another candidate of the third embodiment, only the target layer is the candidate target various processes corresponding to the individual thickening • thinning process or the like extracted by the extracting.

[0101] 本实施例中的图像处理装置中,在存储装置的程序存储单元中,存储着2种类(z^ 2)相似度(确信度)判定程序。 [0101] In this embodiment of an image processing apparatus, the program storage unit in the storage device, stores two kinds (z ^ 2) the degree of similarity (degree of certainty) determining program. 这点与第1实施例下的情况不同。 This is the case in the embodiment differs from the first embodiment. 即,CPU基于z种类的相似度(确信度)判定,对z种类对象进行处理,为了提取增厚•减薄等的目标。 That is, the type of the CPU based on the z similarity (degree of certainty) the determination of the type of the object z processed in order to extract target thickening • thinning the like. 也就是说,对于2种类对象,每个种类生成1个层,分别确定初期候补目标,进行候补目标的变换,提取确定目标。 That is, for the type of the target 2, each generating a type layers, respectively, to determine the initial target candidates, conversion candidates for the target, the extraction determining a target.

[0102]此时,会发生这样的事,即选择某目标作为已知对象物相关初期候补目标(第_ 期候补目标),且选择其他目标作为其他已知对象物相关初期候补目标(第2初期候补目标)。 [0102] At this time, this will happen, i.e., select a relevant target object is known as the initial target candidate (the candidate of the target _), and select another target object as the other known associated initial candidate target (2 the initial target candidates). 这时,关于属于各附带层的附带目标,不管对于不同种类的候补目标中的任何一个,部可以作为增厚•减薄用目标来加以利用。 At this time, on target comes included with each layer belongs, whether a candidate for any of the different types of target, as a thickened portion may be thinned with a goal to • be used. 也就是说,各附带目标即可作为第1候补目标的增厚•减薄用目标,也可以作为第2候补目标的增厚•减薄目标。 In other words, each target can be used as thickening comes first candidate target • thinning with a goal, can be used as thickening second candidate targets • thinning target.

[0103]实施例5: [0103] Example 5:

[0104]以下,就图像识别方法进行说明,该图像识别方法使用了图像识别装置,该图像识别装置包含本发明第5实施例相关的图像处理装置。 [0104] Hereinafter, the image recognition method to be described, the image recognition method using an image recognition apparatus, the image recognition apparatus including an image processing apparatus relating to a fifth embodiment of the present invention. 图I2 (A)是概念性说明本实施例相关图像识别装置500构造的框图。 FIG I2 (A) 500 is a block diagram conceptually illustrating the configuration of the present embodiment related to the image recognition apparatus embodiment. 图I2 〇3)是概念性说明图像识别装置500中存储装置512构成的框图。 FIG 〇3 I2) is a conceptual block diagram illustrating an image recognition apparatus 500 in the storage apparatus 512 thereof. 图像识别装置500是对实施例1中所示图像处理装置中各种图像处理程序等进行添加,进一步附加进行图像识别处理的程序的所形成的。 The image recognition apparatus 500 is the embodiment of the image processing apparatus shown in various image processing program or the like is added 1, formed by the additional program further performs image recognition processing. 也就是说,图像识别装置500包含作为实施例1的图像处理装置的功能,进而利用此功能,再进行图像识别处理。 That is, the image recognition means 500 comprises an image processing apparatus according to an embodiment of the function, and thus the use of this function, then the image recognition process. 为此,关于与图像识别装置500构造中的图像处理装置的共通点,在此省略说明。 To this end, the common point of the image processing apparatus 500 configured in the image recognition apparatus, the description is omitted. 本实施例中,利用图像处理装置所提取的新目标,进行图像识别,可切实并快速的进行图像识别处理。 In this embodiment, the new target using the extracted image processing apparatus, image recognition can be reliably and quickly perform image recognition processing.

[0105]以下,就图像识别装置500的详细构造及动作进行说明。 [0105] The following describes the detailed structure and operation of the image recognition device 500. 如图12(A)所示,图像识别装置500具有CPU510、存储装置512、显示装置513、输入装置514、总线550等,对己拍摄的未知对象物的为特定图像数据,进行已知对象物相关识别处理。 FIG 12 (A), the image recognition device 500 has a CPU 510, a storage device 512, 513, an input device 514, a bus 550 like a display, an unknown object has a particular captured image data, subjected to a known object related to the recognition process. 本实施例中,在构成图像识别装置500的各单元中,输入装置514具有外部数据接收单元514a,可以接收待识别的对象即原图像相关数据、确认是否有识别对象等的对象即未知图像数据等外部图像数据。 Objects of the present embodiment, the respective units constituting the image recognition device 500, input device 514 with an external data receiving unit 514a, may be received to be recognized, i.e., the original image data to confirm whether the recognition target object such as i.e. the unknown image data external image data and the like. 另外CPU510将外部数据接收单元514a接收的各种信息存储于存储装置512的数据存储单元DM 中。 Further the CPU510 in the storage device 512 stores various information data reception unit 514a receives the external data storage unit DM. 通过以上,图像识别装置500分析外部输入的各种图像数据,进行图像识别的处理。 By the above, various image data externally input apparatus 500 analyzes the image recognition, image recognition processing.

[0106]如图12⑻所示,图像识别装置500的存储装置512是有程序存储单元PM和数据存储单元DM构成。 [0106] As shown, the image recognition device 512 storage device 500 is a program storage unit and a data storage unit DM PM configured 12⑻. 程序存储单元PM具有存储各种程序的程序领域。 PM program storage unit that stores various programs having program area. 数据存储单元DM具有存储各种数据的数据领域。 DM data storage unit having a data field for storing various data. 构成要素中,作为被包含于程序存储单元PM的部分,除了具有进行处理对象图即图像的前处理的前处理程序AP外,还具有目标数据生成程序DPX、原图像数据变换程序0PX以及数据比较程序DCX。 Constituent elements, a portion is included in a program storage unit PM, in addition to having a processing target in FIG i.e. outer front handler pretreatment AP image, but also a target data generation program DPX, comparing the original image data conversion program 0PX and data program DCX. 这些各种程序中,目标数据生成程序DPX是根据未知图像数据生成新目标的各种程序集合体,具体来讲,包含第1实施例中说明的层生成程序、增厚•减薄程序等各种程序。 These various programs, data generation program DPX target is unknown image data in accordance with various programs to generate a new target assembly, specifically, a first layer comprising generating program described in Example 1, and other programs thinning thickening • species program. 另外,原图像数据变换程序0PX变换包含应该让识别对象的原本图像的原图像数据,提取该对象原本图像即成为比较对象原本图像的已知图像数据。 Further, the original image data conversion program should allow 0PX recognition target comprises converting original image data of the original image, extracting the object original image becomes a comparison object original image data of an image is known. 数据比较程序DCX比较原图像数据变换程序0PX提取的识别对象数据以及目标数据生成程序DPX 提取的新目标的相关数据,提取各种信息即识别相关数据,包括,新目标中是否包含相当于识别对象数据的数据,如果包含,在图像中的哪一个位置,或者符合的目标个数有几个灯识别对象数据相关信息。 New target data comparison program data DCX comparing the original image data conversion program 0PX the extracted recognition target data and the destination data generation program DPX extraction, i.e. to extract various kinds of information related to the identification data, including, whether to include the new target objects corresponding to the identification the data, if included, which position in the image, or the number of targets in line with the lamp of a few data identifying object information.

[0107]存储装置512作为构成要素中被包含于数据存储单元的装置,除了具有存储未知对象物相关图像信息即未知图像数据的未知图像数据存储单元D0之外,还具有目标相关数据存储单元0DX、识别对象原图像数据存储单元0IX、识别对象数据存储单元TDX、识别相关数据存储单元RDX。 [0107] storage device 512 are included as constituent elements in the apparatus data storage unit, in addition to storing the image information related to an unknown object Unknown Unknown i.e. image data storage unit other than the image data D0, further comprising a target data storage unit 0DX identifying target original image data storing unit 0IX, TDX recognition target data storing unit, the identification data storage unit RDX. 这些各种数据存储单元中,目标相关数据存储单元〇dx显示的是根据未知图像数据生成新目标的各种数据集合体,具体来讲,包含第1实施例中说明的层生成数据存储单元、目标数据存储单元等各种数据存储单元。 These various data storage unit, a target data storage unit 〇dx displays various data based on the image data to generate a new aggregate unknown target, specifically, the data storage unit comprising a layer generating a first embodiment illustrated embodiment, target data storage unit and other data storage unit. 识别对象原图像数据存储单元0IX存储着原图像数据变换程序0PX的变换对象即原图像数据。 Identifying object 0IX original image data storage unit stores the original image data conversion program 0PX conversion object, i.e. the original image data. 另外,识别对象数据存储单元TDX存储着原图像数据变换程序OPX提取的识别对象数据。 Further, the recognition target data storage unit stores TDX recognition target data of the original image data conversion program OPX extracted. 另外,识别相关数据存储单元RDX存储着数据比较程序DCX中提取的识别相关数据。 In addition, identification data related to the identification data storage unit stores data comparison program RDX DCX extracted.

[0108]以下,就图像识别装置500的图像识别方法进行概要性说明。 [0108] Hereinafter, the image recognition method to the image recognition apparatus 500 will be schematically described. 首先,图像识别装置500的CPU510从存储装置5丨2读取(步骤S501)包含于识别对象原图像数据存储单元〇IX中的识别对象的原图像数据(识别对象原图像数据),读取原图像数据变换程序0PX,提取识别对象数据,将该识别对象数据存储于识别对象数据存储单元TDX中(步骤S502)。 First, CPU 510 of the image recognition apparatus 500 comprises storage means 5 Shu 2 read (step S501) from the original image data (original image data recognition target) identifying a target original image data storage unit 〇IX the recognition target, the original reading 0px image data conversion program, extracts the recognition target data, the identification data stored in the object data storing unit TDX identified object (step S502). 其次,CPU510 恰当读取目标数据生成程序DPX等,读取未知图像数据的信息(步骤S503),提取各种目标(步骤S504)。 Next, CPU 510 reads the target data generation program appropriate DPX like, reads the image information of the unknown data (step S503), extracts various target (step S504). 关于步骤S503及S504的处理,可应用上述各实施例中的各种分割处理方法。 Processing regarding S504 and step S503, the division processing method of the above-described various embodiments may be applied. 其次,CPU510读取数据比较程序DCX,比较步骤S5〇2中存储的成为比较对象的原图像的已知图像数据即识别对象数据,和步骤S503以及S504中提取的未知图像数据相关目标的数据,识别相当于识别对象数据的数据是否包含于新目标中(步骤S5〇5)。 Next, CPU 510 reads the data comparison program DCX, the comparing step S5〇2 be stored in a known image data of the original image, i.e., comparison target recognition target data, and steps S503 S504 and the data extracted from the unknown image data related to the target, identifying whether data corresponding to the object identification data contained in the new target (step S5〇5). 即,基于根据未知图像数据提取的新目标,至少判定有无已知对象物。 That is, based on the new target image data extracted from the unknown, it is determined whether or at least a known object. 最后,CPU510将步骤S505中取得的各种信息作为识别相关数据存储于识别相关数据存储单元RDX中(步骤S506)。 Finally, CPU510 various information acquired in step S505 as the identification data stored in identification data storage unit associated RDX (step S506).

[0109]以下,参照图I3就基于上述步骤S5〇5的比较处理的图像识别方法相关第1例进行说明。 [0109] Hereinafter, with reference to FIG I3 will be described first example of related image recognition method of the above-described comparison processing of step S5〇5 based. 如图所示,在此,作为图像识别,从1个对象即全体图像PI提取作为识别对象的识别对象数据AS (或者识别对象数据AS1)。 As shown, in this case, image recognition, i.e., the entire image PI extracts the recognition target object data identification AS (identified object or data AS1) from an object. 此处的识别对象数据AS是从全体图像PI的指定轮廓提取的部分图像即1种目标。 AS identifying object data is herein designated partial image extracted from the whole contour, i.e. one kind of image PI target. 关于从识别对象数据AS的轮廓提取的方法,可应用手动、自动等各种方法。 Method AS on the recognition target data from the contour extraction, can be applied manually, automatically, and other methods. CPU510对于指定的识别对象数据AS,从图中1点短线所示多个候补目标E0 (E01, E02,…,E010”〇中选择最适合的目标(图示例子中候补目标E03)作为对应的目标X0。关于所选择的目标X0,比如,不仅仅限于1个,有时也可选择多个。另外,关于所选择目标X0的位置的信息,关于识别对象,根据需要,可恰当选择应该获取的目标。此时,也就是说,通过比较(对比)各目标,进行图像识别。关于对比方法,可应用已知的各种方法。 CPU510 for a given recognition target data AS,, ..., E010 square select the most suitable target from a plurality of candidates 1:00 FIG target E0 (E01, E02 shown in dashes "(E03 target candidate in the illustrated example) corresponding to a target X0. X0 on the selected target, for example, is not limited to one, and sometimes a plurality of select Further, information regarding the position of the selected target X0, concerning the identification target, if necessary, can be appropriately selected to be acquired target. in this case, that is, by comparison (comparative) of each target, image recognition. Contrasting method, various known methods may be applied.

[0110]以下,参考图14,就基于上述步骤S505中比较处理的图像识别方法相关第2例进行说明。 [0110] Hereinafter, with reference to FIG. 14, it is related to the second embodiment will be explained the image recognition method of the above-described process in step S505 based on the comparison. 图13中,将识别对象数据AS作为从原图像数据即整体图像PI提取的目标,在识别处理中,也就是说对各目标进行比较。 13, the data to be recognized as a target AS PI i.e., the entire image from the original image data extracted in the recognition process, that is to say for each comparison target. 对此,如图所示,指定全体图像PI的领域,通过与此领域中特定的图像进行模式匹配,选择目标X0。 In this regard, as shown in FIG PI specified whole image field by performing pattern matching with a specific image in this field, select the target X0. 此时,仅通过指定领域DL,可进行步骤S505的识别处理。 In this case, by designating only the DL field, recognition processing may be performed in step S505 is. 另一方面,因包含对象即图像部分的背景部分,在识别中,多少会产生些噪音。 On the other hand, due to the background portion containing the object image portion i.e., in recognition, much more noise is generated.

[0111]以下,参照图15,就基于上述步骤S505中的比较处理的图像识别方法相关第3例进行说明。 [0111] Hereinafter, with reference to FIG. 15, on the image recognition method of the above-described comparison processing in step S505 based on the correlation of the third embodiment will be explained. 第3例是第1例的应用,相对于第1例扑捉1个目标,本例中,基于多个目标的相对组合,进行识别。 The third embodiment is an application of the first embodiment with respect to the first embodiment to capture a target, in this embodiment, a combination of a plurality of relative target-based identification. 图15㈧〜15⑹中,图15(A)是识别对象即图像相关典型图像样本。 FIG 15㈧~15⑹ in FIG. 15 (A) is an identified object image that is related to a typical image samples. 在此,全体图像中,存在3种类的对象物的数据图像即识别对象数据AS1、AS2、AS3。 Here, all the image data of the image there are three types, i.e., the object recognition target data AS1, AS2, AS3. 如图15 (B)所示, 这些识别对象具有箭头AR1〜AR3所示方向与位置关系。 FIG. 15 (B), these identification objects having a positional relationship shown in the arrow direction AR1~AR3. 即,在此,识别对应3个识别对象数据AS1、AS2、AS3的3个目标是否存在图15⑻所示的位置关系。 That is, in this case, the identification data corresponding to three objects identified AS1, AS2, AS3 three target positional relationship shown in FIG 15⑻ exists. 具体来讲,为了把握上述位置关系,规定图15 (C)所示的阴影部分MK和根据阴影部分MK提取的领域AR1〜AR3,在各领域AR1〜AR3中,识别是否存在对应识别对象数据AS1、AS2、AS3的目标。 Specifically, in order to grasp the positional relation, a predetermined FIG. 15 (C) MK hatched portion shown hatched portion AR1~AR3 art and according to the extracted MK, AR1~AR3 in various fields, identifying whether there is a corresponding identification target data AS1 , AS2, AS3 goal. 图15⑼〜图15⑹是针对未知图像,显示上述识别动作的数据。 FIG 15⑼~ FIG 15⑹ knowing that image, the display data of the identification operation. 首先,如图15⑼所示,对应领域AR 1〜AR3的第1领域〜第3领域UA1〜UA3中,在第1领域UA1中,判定是否存在符合识别对象数据AS1 (参照图15 (C))的目标。 First, as shown in FIG 15⑼, AR corresponding to the first field to the third field of the art 1~AR3 UA1~UA3, the UA1 in the first field, determines whether there is target data corresponding to the identification ASl (see FIG. 15 (C)) The goal. 在此,如图15⑻所示,若存在符合的目标X01,接着,在第2领域UA2中,判定是否存在符合识别对象数据AS2 (参照图15⑹)的目标。 Here, as shown in FIG 15⑻, X01 matching if the target is present, then, in the second field UA2, it is determined whether the target objects corresponding to the identification data the AS2 (see FIG 15⑹) a. 进而,如图15⑻所示,判定是否存在符合识别对象数据AS3 (参照图15⑹)的目标。 Further, as shown in FIG 15⑻, it is determined whether the target object data corresponding to the identification AS3 (see FIG 15⑹) exists. 如图15⑹所示,存在符合目标x〇3时,判定符合识别对象即图像。 As shown in FIG 15⑹, in line with the presence of the target x〇3, i.e., it is determined in line with the recognition target image.

[Q112]以下,就上述第3例中图像识别装置的图像识别顺序进行说明。 [Q112] hereinafter be described in order of the above image recognition in the image recognition apparatus of the third example. 首先,CPU510基于识别对象原图像相关数据,决定对象领域(第1领域UA1〜第3领域UA3)处理(步骤S601)。 First, CPU 510 of the original image based on the object identification data, decision objects field (first field of 3 UA1~ art UA3) processing (step S601). 其次,CPU510恰当读取识别对象数据,进行第1领域UA1的识别处理(步骤S6〇2)。 Next, CPU 510 appropriately reads the identification object data, a recognition process (step S6〇2) UA1 of the first field. CPU510判断第1领域UA1中存在对应目标的数据(步骤Se〇3: Yes),在临时保存第1领域UA1中的识别结果(步骤S604),开始第2领域UA2的识别处理(步骤S6〇5) XPU判断第2领域UA2中存在对应目标的数据(步骤S6〇6:Yes),临时保存第2领域UA2中的识别结果(步骤Se〇7),开始第3领域UA3 的识别(步骤S6〇8) XPUHO判断第3领域UA3中存在对应目标的数据(步骤S609:Yes),判定有符合的数据,保存第1领域〜第3领域UA1〜UA3的各识别结果(步骤S610)。 There is a corresponding target data CPU510 determines whether the first field of UA1 (step Se〇3: Yes), the recognition result (step S604) temporarily stored in the first field of UA1, the second field recognition process starts UA2 (step S6〇5 ) (step S6〇6 XPU determines the presence of the corresponding object in the second field UA2: Yes), the provisional identification result stored (step 2 Se〇7 UA2 in the art), identifying the start of the third field UA3 (step S6〇 data (step S609 the corresponding target is present in 8) XPUHO UA3 determined in the third field: Yes), determines that there is matching data, - storage of the first field in each of the third field UA1~UA3 recognition result (step S610). 另一方面,上述中,判断各领域UA1〜UA3的识别处理(步骤S6〇2、S6〇5、S606)中,判断其中任何一个不存在相符数据(步骤S6O3: No、步骤S605: No、步骤S6〇9:No),保存不存在符合目标这样的信息(步骤S611)。 On the other hand, in the above, the identification process is determined in various fields UA1~UA3 (step S6〇2, S6〇5, S606), it is determined which corresponds to one of any data (step S6O3 absent: No, Step S605: No, Step S6〇9: no), save the absence of such information in line with the target (step S611).

[0113]根据本实施例相关图像识别装置,在提取的多个分割图像中,可更加准确捕捉对象物,以通常模式匹配中无法捕捉的精度,对目标进行识别。 [0113] The accuracy of the image recognition apparatus according to the present embodiment related embodiment, the plurality of divided images extracted, the object can be more accurately capture, matching in the normal mode can not be captured, the target identification.

[0114]实施例6: [0114] Example 6:

[0115]以下就使用包含本发明的第6实施例相关图像处理装置的图像分类装置的图像分类方法进行说明。 Image classification method according to an image classification apparatus related image processing apparatus of the sixth embodiment [0115] of the present invention containing the following will be described. 本实施例中图像分类装置的构成与图12(A)和图12(B)所示即第5实施例的图像识别装置构成相同,在此省略详细图解及说明,根据需要,请适当参照图12 (A)等。 FIG configuration example image classification apparatus of the present embodiment 12 (A) and 12 (B) that is the same as the image recognition apparatus constituting a fifth embodiment, in this detailed illustration and description thereof will be omitted, if desired, please see Figs 12 (A) and the like. [0116]图16㈧是本实施例相关图像分类装置的处理对象即未知图像UI的例子。 [0116] FIG 16㈧ processed according to the present embodiment is related to an image classification apparatus embodiment that is an example of a UI image is unknown. 在此,所谓图像分类装置中的图像分类是指在图16(A)那样的未知图像UI中,特别指定多个种类已知对象物,即多个种类己知对象物的图像有多少个,在哪里。 Here, the image classification apparatus called the image classification means (A) as the unknown image UI, the specified object in a plurality of known type 16, i.e., a plurality of image types known how many of the object, where is it. 比如,对于4种类已知对象物, 进行特别指定。 For example, a known type for the object 4, for specified. 图16 (B)〜16 (E)是显示进行上述分类所需要的知识数据DI1〜DI4。 FIG. 16 (B) ~16 (E) is a knowledge data required DI1~DI4 above classification. 具体来讲,各知识数据DI1〜DI4分别是包含多个关于1个己知对象数据的图像数据。 Specifically, each of the knowledge data are DI1~DI4 comprising a plurality of image data on a known object data. 图像分类装置基于这些知识数据DI1〜DI4,判断未知的图像属于4个知识数据DI1〜DI4中所显示的已知对象中的哪一个,或者不属于任何一个。 Image classification based on these knowledge data DI1~DI4 apparatus, which determines the unknown image belongs to a known object 4 DI1~DI4 knowledge data in the displayed or does not belong to any one. 知识数据DI1〜DI4作为识别对象数据,被存储于图12 (B)的识别对象数据存储单元TDX中。 DI1~DI4 knowledge data as the recognition target data, is in FIG. 12 (B) storing the identification target in the data storing unit TDX.

[0117]以下,就图像分类装置的图像分类方法概要进行说明。 [0117] Here, the image classification method on a schematic image classification device will be described. 如所述,关于图像分类装置的构成,借用图12 (A)。 As mentioned, the configuration of an image classification apparatus, borrowing FIG 12 (A). 首先,CPU510设定多个种类(此处为4个种类)对象物相关知识数据DI1〜DI4(步骤S701)。 First, CPU510 sets a plurality of types (here, four types) of the object knowledge data DI1~DI4 (step S701). 即,根据未知图像,决定应分类的己知对象。 That is, according to the unknown image is determined to be the object classification known. 其次,CPU510读取未知图像数据即未知图像UI的信息(步骤S702),提取各种目标(步骤S703)。 Then, CPU510 reads the image data that is unknown information unknown image UI (step S702), extracts various target (step S703). 关于步骤S702以及步^S703的处理,可适用上述各种实施例中所示的各种分割处理方法。 About ^ S703 step S702 and step processing, applied the above-described various embodiments shown in the various segmentation approach. 另外,此时,比如可以采用对应实施例1中设定的对象物的分割。 In this case, the object may be divided for example in Example 1 using a set corresponding to the embodiment. 其次,CPU510读取多个种类对象物相关知识数据DI1〜DI4(步骤S703),就被提取的各种目标进行数据比较(步骤S7〇4)。 Then, CPU510 reads the plurality of related knowledge data type of the object DI1~DI4 (step S703), the data on various objects are extracted (step S7〇4). 也就是说,CPU510判定是否存在符合4个对象物的目标,或者判定哪个目标与哪个对象物相符(步骤S705〜步骤5706) 。 In other words, CPU510 determines whether there is a target in line 4 of the object, or to determine which target with which the object is consistent (S705~ Step 5706). 即,在步骤S705中,确认有无相符数据,判定有时(步骤S705: Yes),关于相符目标,进行数据保存(步骤S706)。 That is, in step S705, the data to confirm whether the match, sometimes determined (step S705: Yes), in line on the target, data to be saved (step S706). 判断无时(步骤S705:N〇),保存没有相符目标这样的信息(步骤5707) 。 (Step S705: N〇) is determined not to save such information did not match the target (step 5707). 对于各种目标的目标进行上述数据保存的相关处理,可以分类图像分类装置的未知图像中多个种类的对象物。 Correlation processing for the target data stored in the above-described various targets, can classify an unknown image classification apparatus of the plurality of object types.

[0118]根据本实施例相关图像分类装置,比如,在1个图像内存在不同物体(多个种类的对象物)时,可把握这些不同物体的个数以及分布等。 [0118] The image classification apparatus according to the present embodiment related embodiment, for example, in an image memory at different objects (a plurality of types of object), which can grasp the number and the distribution of different objects, and the like. 更具体来讲,比如,关于血液的图像, 可以更加正确进行血液中各成分以怎样的比例存在等这样的分析。 More specifically, for example, the image on the blood, such analysis can be more blood components in what proportions, etc. correctly. 另外,关于某对象物(血液中的成分),可进行正常物与异常物以怎样的比例存在等这样的分析。 Further, on an object (blood component), such an analysis was normal and abnormal in what proportions thereof and the like may be performed.

Claims (10)

1. 一种图像处理装置,变换从拍摄的未知对象物的未知图像数据所提取的目标,提取对应已知对象物的新目标,其特征在于,包括以下单元: 层生成单元,生成由包含根据未知图像数据所提取的目标的分割图像集合体构成的层; 初期候补目标决定单元,根据所述层生成单元生成的层中所包含的目标,和该目标进行变换后而得到的目标,决定其是否作为初期候补目标; 候补目标变换单元,利用所述层生成单元生成的层中所包含的所述初期候补目标之外的目标,变换初级候补目标,生成新的候补目标; 相似度测定单元,测定所述目标与已知对象物的相似度; 目标确定单元,根据所述相似度测定单元测定的相似度测定结果,决定通过所述候补目标变换单元的变换所生成的新的候补目标和变换前的候补目标中所选择的目标中的哪个作为确定目标进行保 An image processing apparatus, converting the unknown image data from an unknown target object photographed extracted, extracts the corresponding new target object is known, wherein the unit comprising: layer generating means is generated by comprising the unknown aggregate layer divided image data of the extracted image object configuration; initial candidate target determining unit, according to the layer generating unit generates the target layer contained, and after the object obtained by converting a target, determine their whether a candidate as an initial goal; target candidate converting means, using the target than the initial target of the layers candidate generating unit included in the converted primary target candidate generating a new candidate target; similarity measurement unit, Determination of the degree of similarity with known target object; target determining unit, the similarity measurement unit measuring a similarity measurement result, and determining a new candidate target generated by the conversion candidate of the conversion according to target transform unit target candidates in the pre-selected targets in which Paul performed as a determination target ; 其中,所述初期候补目标决定单元基于所述相似度测定单元的测定结果,决定所述初期候补目标;所述目标确定单元基于所述相似度测定单元所测定的相似度,基于事先规定的阈值,决定所述目标是否为应该保持的确定目标; 所述目标确定单元基于所述相似度测定单元测定的相似度测定结果,比较通过所述候补目标变换单元变换而生成的新的候补目标和变换前的候补目标,判断是重复所述目标变换单元的变换动作还是结束该动作,并将该判断结果所得到的候补目标作为所述确定目标进行保存; 所述初期候补目标决定单元将所述层生成单元生成的各种目标中超过所述相似度测定单元的相似度阈值的目标作为初期候补目标,所述相似度测定单元将提取的目标与所述已知对象物的相似度进行数值化的确定度作为判定的基准。 ; Wherein the initial candidate target determining unit based on the measurement result of the similarity determination unit determines the initial candidate target; the target determination unit determined based on the similarity of the similarity measurement unit, based on prespecified threshold, the target determines whether the determination target should remain; new candidate targets the target determining unit based on the measurement result of the measurement unit measures the similarity of the similarity, comparison of the candidate target generated by the transform unit and candidate target before conversion, the judgment is repeated conversion operation target transform unit or the operation is finished, and the determination result obtained in the determination target candidate as a target to save; the initial candidate decision unit sets the target various target layer generating unit in the similarity exceeds a similarity threshold measurement unit as an initial target a target candidate, the similarity determination unit value with the extracted target object similarity of the known degree of certainty as a reference determination.
2. 根据权利要求1所述的图像处理装置,其特征在于:所述层生成单元有第1层生成单元即生成由包含所述初期候补目标的目标群构成的层,和第2层生成单元即生成仅仅由不包含所述初期候补目标的目标群构成的特别层;所述第2层生成单元生成仅由附带目标构成的附带层,将其作为所述特别层,该附带目标在所述未知图像数据提取目标时的图像处理中被从构成所述候补层的各种目标中排除,而在所述候补目标变换单元的变换中,被附带于其他目标。 The image processing apparatus according to claim 1, wherein: said first layer with a layer generating unit generating means which generates a layer composed of a group including the initial candidate target object, and the second layer generating unit i.e. generation layer consisting only of particular target group does not include the initial candidate object; with the layer of the second layer generating unit generates a target composed of only incidental, as it is the particular layer, which comes in the target image processing of image data to extract the unknown target is excluded from the candidate constituting various target layer, and in transforming the target conversion candidate unit is incidental to the other objectives.
3. 根据权利要求2所述的图像处理装置,其特征在于:所述层生成单元还具有第3层生成单元,即生成其他候补层,该其他候补层由与上述己知对象物不同的其他己知对象物中提取的其他候补目标相关分割图像集合体构成的,所述候补目标变换单元在生成新的候补目标时,将所述其他候补层中包含的所述其他候补目标从变换动作对象中排除。 The image processing apparatus according to claim 2, wherein: said generating means further includes a layer of a third layer generating unit which generates a candidate other layer, the other layer is made different from the above candidate known object other other known target object candidate extracted divided images related to the aggregate of the target candidate converting means when generating a new target candidate, the candidate of the other layer comprises another candidate target objects from the transform operation excluded.
4. 根据权利要求2所述的图像处理装置,其特征在于:所述层生成单元生成的层是具有阶层性构造的层群,含相对的上方层与下方层,处于上方的上方层包含父目标,该父目标将构成所述下方层的多个目标作为子目标包含在内部;所述初期候补目标决定单元在所述层生成单元生成的所述阶层性构造的层群中,根据属于下方层的目标,决定所述初期候补目标;所述候补目标变换单元将所述初期候补目标以及附着其他目标,生成新的候补目标;所述初期候补目标决定单元在所述层生成单元生成的所述阶层性构造的层群中,根据属于所述上方层的目标,决定所述初期候补目标;所述候补目标变换单元对于所述初期候补目标, 去除所述初期候补目标中被包含的子目标的一部分,生成新的候补目标。 4. The image processing apparatus according to claim 2, wherein: the layer generation unit layer is a layer having a hierarchical configuration of the group containing a relatively upper layer and the lower layer, located above the top of the layer comprising the parent target, the parent object constituting the underlying layer as the plurality of target sub-targets contained therein; the initial candidate target determining unit generates the hierarchical layer group configuration unit generated in the layer, according to the lower part of target layer determines the initial candidate target; the target candidate converting means and the initial target candidate attached to other objects, to generate new candidate target; the initial candidate decision unit generating the target cells generated in the layer said hierarchical group layer configuration, according to the above belonging to the target layer, determines the initial candidate target; the target candidate converting means to the initial candidate target, removing the initial subgoal to be included in the candidate target part of a new generation of candidate targets.
5.根据权利要求4所述的图像处理装置,其特征在于:所述层生成单元对于既存分割图像,测定各分割图像尺寸,比较测定的该分割图像的图像数据尺寸和对应所述己知对象物的图像数据尺寸,根据需要,再分割该分割图像,生成基于新目标的层。 The image processing apparatus according to claim 4, wherein: said means for generating an existing layer divided images, each divided image size determination, the size of the image data of the divided image is determined by comparison to the known objects corresponding the image data size thereof, if necessary, re-division of the divided images, a new object based on generating layer.
6.—种变换根据拍摄的未知对象物的未知图像数据提取的目标,提取对应已知对象物的新目标的图像处理方法,具有以下工序: 层生成工序,生成由包含从所述未知图像数据提取的目标的分割图像集合体构成的层; 初期候补目标决定工序,根据被包含于所述层生成工序生成的层中的目标,决定作为初期候补被变换而成的处理对象即初期候补目标; 候补目标变换工序,利用所述层生成工序生成的层中包含的所述初期候补目标以外的目标,通过变换所述初期候补目标,生成新的候补目标; 相似度测定工序,测定所述目标与已知对象物的相似度; 目标确定工序,根据所述相似度测定工序测定的相似度测定结果,决定通过所述候补目标变换工序的变换所生成的新的候补目标和变换前的候补目标中所选择的目标中的哪个作为确定目标进行保存; 其中,所 6.- The transformed target species unknown unknown object image data captured extracted, extracting an image processing method corresponding to the new target object is known, the steps of: generating a layer step of generating from the image data generated by the unknown comprising divided image layer extracted target assembly composed of; the initial candidate target determination step, is included in the layer in accordance with step of generating the target generation layer, it determines a processing target is converted from the beginning as an initial candidate i.e. a candidate target; candidate target conversion step, using a target other than the target layer is the initial candidate generating step of generating the layer contains, by transforming the initial target candidates, generating a new candidate target; similarity determination step of determining the target and similarity known object; target determining step, the step of measuring the similarity measurement results, determine the new candidate objects and candidate target generated before conversion candidate by transforming the target conversion step according to the similarity measurement which is saved as a determination target in the selected target; wherein the 初期候补目标决定工序基于所述相似度测定工序的测定结果,决定所述初期候补目标;所述目标确定工序基于所述相似度测定工序所测定的相似度,基于事先规定的阈值,决定所述目标是否为应该保持的确定目标; 所述目标确定工序基于所述相似度测定工序测定的相似度测定结果,比较通过所述候补目标变换工序变换而生成的新的候补目标和变换前的候补目标,判断是重复所述目标变换工序的变换动作还是结束该动作,并将该判断结果所得到的候补目标作为所述确定目标进行保存; 所述初期候补目标决定工序将所述层生成工序生成的各种目标中超过所述相似度测定工序的相似度阈值的目标作为初期候补目标,所述相似度测定工序将提取的目标与所述已知对象物的相似度进行数值化的确定度作为判定的基准。 Initial candidate destination determination step based on the measurement result of the similarity determination step determines the initial candidate target; the target similarity determining step of the similarity measurement step determined based on a predetermined threshold value in advance, determines the should be determined whether the target objectives; and determining said target step similarity measurement step measurement results, by comparing the candidate target before a new candidate of the candidate target conversion step of converting the target generated and converted based on the similarity measurement determining a target conversion step of repeating the converting operation, or the operation is finished, and the determination result obtained in the determination target candidate as a target to save; target determining step of the initial candidate generating step of generating the layer of various target exceeds the similarity determination target similarity threshold step as an initial target candidate, the similarity of the extracted measuring step of determining a target value for the degree of similarity of the known object is determined as benchmark.
7.—种图像识别装置,其包含权利要求1至5中任一项所述图像处理装置,该图像识别装置根据该图像处理装置从所述未知图像数据提取的新目标,进行有无已知对象物的判定。 7.- kinds of image recognition apparatus, comprising an image processing according to claim 1 to 5, in any apparatus, the image recognition means new and unknown target image data extracted from the image processing apparatus according to the presence or absence of known determining the object.
8.—种根据拍摄的未知对象物的未知图像数据,进行已知对象物相关识别的图像识别方法,具有判定工序图像识别方法,包含权利要求6中所述的图像处理方法,基于根据该图像处理方法从所述未知图像数据提取的新目标,进行有无已知对象物的判定。 8.- The species unknown unknown object image data captured, the image recognition method known in the relevant object recognition, image recognition method having a step of determining, comprising image processing method according to claim 6, on the basis of the image the method of determining unknown process from the new target image data extracted, presence or absence of a known object.
9.一种根据拍摄的未知对象物的未知图像数据,进行已知对象物相关识别的图像分类装置,包含权利要求1至5中任一项所述的图像处理装置,基于根据该图像处理装置从所述未知图像数据提取的新目标,进行己知对象物的分类。 An unknown image data in accordance with an unknown object photographed, an image classification apparatus to identify the relevant object is known, comprising an image processing apparatus as claimed in claim 1 to 5 according to any of the image processing apparatus according to the basis the new target from an unknown image data extracted, known classifying the object.
10.—种根据拍摄的未知对象物的未知图像数据,进行己知对象物识别的图像分类方法,包含权利要求6所述的图像处理方法,基于根据该方法从所述未知图像数据提取的新目标,进行已知对象物的分类。 10.- The species unknown unknown image data captured object, an image classification method known object recognition, comprising the image processing method as claimed in claim 6, according to the method based on the extracted new image data from said unknown target, classification known object.
CN201410102668.7A 2014-03-19 2014-03-19 Image processing, image recognition and image classification apparatus and method CN103824085B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410102668.7A CN103824085B (en) 2014-03-19 2014-03-19 Image processing, image recognition and image classification apparatus and method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201410102668.7A CN103824085B (en) 2014-03-19 2014-03-19 Image processing, image recognition and image classification apparatus and method
JP2015050161A JP2015179511A (en) 2014-03-19 2015-03-13 Image processing device and method, image recognition device and method, and image classification device and method

Publications (2)

Publication Number Publication Date
CN103824085A CN103824085A (en) 2014-05-28
CN103824085B true CN103824085B (en) 2018-02-02

Family

ID=50759137

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410102668.7A CN103824085B (en) 2014-03-19 2014-03-19 Image processing, image recognition and image classification apparatus and method

Country Status (2)

Country Link
JP (1) JP2015179511A (en)
CN (1) CN103824085B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101320426A (en) * 2007-06-06 2008-12-10 夏普株式会社 Image processing device and method, image forming device and image processing system
CN103034844A (en) * 2012-12-10 2013-04-10 广东图图搜网络科技有限公司 Image identification method and device
CN103268496A (en) * 2013-06-08 2013-08-28 中国人民解放军国防科学技术大学 Target identification method of SAR (synthetic aperture radar) images

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8224078B2 (en) * 2000-11-06 2012-07-17 Nant Holdings Ip, Llc Image capture and identification system and process

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101320426A (en) * 2007-06-06 2008-12-10 夏普株式会社 Image processing device and method, image forming device and image processing system
CN103034844A (en) * 2012-12-10 2013-04-10 广东图图搜网络科技有限公司 Image identification method and device
CN103268496A (en) * 2013-06-08 2013-08-28 中国人民解放军国防科学技术大学 Target identification method of SAR (synthetic aperture radar) images

Also Published As

Publication number Publication date
CN103824085A (en) 2014-05-28
JP2015179511A (en) 2015-10-08

Similar Documents

Publication Publication Date Title
KR101457284B1 (en) Methods and apparatuses for facilitating content-based image retrieval
JP4990383B2 (en) METHOD representation of the image group, the search method of an image group, apparatus, computer readable storage media and computer systems
CN101536035B (en) Image recognition method, image recognition device, and image recognition program
US20160148079A1 (en) Object detection using cascaded convolutional neural networks
CN102105901B (en) Annotating images
US8712154B2 (en) Image processing system and method
CN101842809B (en) Information processing apparatus and information processing method
JP5449119B2 (en) Method and system for determining the disparity search range in the stereoscopic
EP3204888A1 (en) Spatial pyramid pooling networks for image processing
CN103544506A (en) Method and device for classifying images on basis of convolutional neural network
Tombari et al. Full-search-equivalent pattern matching with incremental dissimilarity approximations
CN101681501A (en) Image processing apparatus, method, and storage medium
JP5282658B2 (en) Image learning, automatic annotation, the retrieval method and apparatus
CN105981008A (en) Learning deep face representation
JP4545641B2 (en) Similar image retrieval method, image retrieval, image retrieval program, and a recording medium
CN102566981B (en) Parallel processing machine learning decision tree training
US9965719B2 (en) Subcategory-aware convolutional neural networks for object detection
Chli et al. Active matching for visual tracking
KR20160083127A (en) Method and system for face image recognition
CN102857686B (en) The method of image processing apparatus, controls the image processing apparatus
US20120051634A1 (en) System for Computationally Quantifying Similarities Between Images
JP4877374B2 (en) An image processing apparatus and program
CN103259962B (en) An object tracking method and associated apparatus
JP2008262331A (en) Object tracking device and object tracking method
JP5414416B2 (en) Information processing apparatus and method

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
GR01